Daily Papers

Recent developments in Large Language Models (LLMs) have manifested significant advancements. To facilitate safeguards against malicious exploitation, a body of research has concentrated on aligning LLMs with human preferences and inhibiting their generation of inappropriate content. Unfortunately, such alignments are often vulnerable: fine-tuning with a minimal amount of harmful data can easily unalign the target LLM. While being effective, such fine-tuning-based unalignment approaches also have their own limitations: (1) non-stealthiness, after fine-tuning, safety audits or red-teaming can easily expose the potential weaknesses of the unaligned models, thereby precluding their release/use. (2) non-persistence, the unaligned LLMs can be easily repaired through re-alignment, i.e., fine-tuning again with aligned data points. In this work, we show that it is possible to conduct stealthy and persistent unalignment on large language models via backdoor injections. We also provide a novel understanding on the relationship between the backdoor persistence and the activation pattern and further provide guidelines for potential trigger design. Through extensive experiments, we demonstrate that our proposed stealthy and persistent unalignment can successfully pass the safety evaluation while maintaining strong persistence against re-alignment defense.

Large Reasoning Models possess remarkable capabilities for self-correction in general domain; however, they frequently struggle to recover from unsafe reasoning trajectories under adversarial attacks. Existing alignment methods attempt to mitigate this vulnerability by fine-tuning the model on expert data including reflection traces or adversarial prefixes. Crucially, these approaches are often hindered by static training data which inevitably deviate from model's dynamic, on-policy reasoning traces, resulting in model hardly covering its vast generation space and learning to recover from its own failures. To bridge this gap, we propose Self-ReSET, a pure reinforcement learning framework designed to equip LRMs with the intrinsic capacity to recover from their own safety error trajectories, which are subsequently reused as an initial state for reinforcement learning. Extensive experiments across various LRMs and benchmarks demonstrate that Self-ReSET significantly enhances robustness against adversarial attacks especially out-of-distribution (OOD) jailbreak prompts while maintaining general utility, along with efficient data utilization. Further analysis reveals that our method effectively fosters self-recovery patterns, enabling models to better identify and recover from unsafe intermediate error states back to benign paths. Our codes and data are available at https://github.com/Ing1024/Self-ReSET.

Large reasoning models (LRMs) "think" by generating structured chain-of-thought (CoT) before producing a final answer, yet they still lack the ability to reason critically about safety alignment and are easily biased when a flawed premise is injected into their thought process. We propose RECAP (Robust Safety Alignment via Counter-Aligned Prefilling), a principled reinforcement learning (RL) method for post-training that explicitly teaches models to override flawed reasoning trajectories and reroute to safe and helpful responses. RECAP trains on a mixture of synthetically generated counter-aligned CoT prefills and standard prompts, requires no additional training cost or modifications beyond vanilla reinforcement learning from human feedback (RLHF), and substantially improves safety and jailbreak robustness, reduces overrefusal, and preserves core reasoning capability -- all while maintaining inference token budget. Extensive analysis shows that RECAP-trained models engage in self-reflection more frequently and remain robust under adaptive attacks, preserving safety even after repeated attempts to override their reasoning.

Large Language Models (LLMs) have enabled the development of powerful agentic systems capable of automating complex workflows across various fields. However, these systems are highly vulnerable to indirect prompt injection attacks, where malicious instructions embedded in external data can hijack agent behavior. In this work, we present ReasAlign, a model-level solution to improve safety alignment against indirect prompt injection attacks. The core idea of ReasAlign is to incorporate structured reasoning steps to analyze user queries, detect conflicting instructions, and preserve the continuity of the user's intended tasks to defend against indirect injection attacks. To further ensure reasoning logic and accuracy, we introduce a test-time scaling mechanism with a preference-optimized judge model that scores reasoning steps and selects the best trajectory. Comprehensive evaluations across various benchmarks show that ReasAlign maintains utility comparable to an undefended model while consistently outperforming Meta SecAlign, the strongest prior guardrail. On the representative open-ended CyberSecEval2 benchmark, which includes multiple prompt-injected tasks, ReasAlign achieves 94.6% utility and only 3.6% ASR, far surpassing the state-of-the-art defensive model of Meta SecAlign (56.4% utility and 74.4% ASR). These results demonstrate that ReasAlign achieves the best trade-off between security and utility, establishing a robust and practical defense against prompt injection attacks in real-world agentic systems. Our code and experimental results could be found at https://github.com/leolee99/ReasAlign.

Ensuring robust safety alignment is crucial for Large Language Models (LLMs), yet existing defenses often lag behind evolving adversarial attacks due to their reliance on static, pre-collected data distributions. In this paper, we introduce MAGIC, a novel multi-turn multi-agent reinforcement learning framework that formulates LLM safety alignment as an adversarial asymmetric game. Specifically, an attacker agent learns to iteratively rewrite original queries into deceptive prompts, while a defender agent simultaneously optimizes its policy to recognize and refuse such inputs. This dynamic process triggers a co-evolution, where the attacker's ever-changing strategies continuously uncover long-tail vulnerabilities, driving the defender to generalize to unseen attack patterns. Remarkably, we observe that the attacker, endowed with initial reasoning ability, evolves novel, previously unseen combinatorial strategies through iterative RL training, underscoring our method's substantial potential. Theoretically, we provide insights into a more robust game equilibrium and derive safety guarantees. Extensive experiments validate our framework's effectiveness, demonstrating superior defense success rates without compromising the helpfulness of the model. Our code is available at https://github.com/BattleWen/MAGIC.

Large Language Models (LLMs) have achieved remarkable capabilities but remain vulnerable to adversarial ``jailbreak'' attacks designed to bypass safety guardrails. Current safety alignment methods depend heavily on static external red teaming, utilizing fixed defense prompts or pre-collected adversarial datasets. This leads to a rigid defense that overfits known patterns and fails to generalize to novel, sophisticated threats. To address this critical limitation, we propose empowering the model to be its own red teamer, capable of achieving autonomous and evolving adversarial attacks. Specifically, we introduce Safety Self- Play (SSP), a system that utilizes a single LLM to act concurrently as both the Attacker (generating jailbreaks) and the Defender (refusing harmful requests) within a unified Reinforcement Learning (RL) loop, dynamically evolving attack strategies to uncover vulnerabilities while simultaneously strengthening defense mechanisms. To ensure the Defender effectively addresses critical safety issues during the self-play, we introduce an advanced Reflective Experience Replay Mechanism, which uses an experience pool accumulated throughout the process. The mechanism employs a Upper Confidence Bound (UCB) sampling strategy to focus on failure cases with low rewards, helping the model learn from past hard mistakes while balancing exploration and exploitation. Extensive experiments demonstrate that our SSP approach autonomously evolves robust defense capabilities, significantly outperforming baselines trained on static adversarial datasets and establishing a new benchmark for proactive safety alignment.

Fine-tuning large language models (LLMs) for telecom tasks and datasets is a common practice to adapt general-purpose models to the telecom domain. However, little attention has been paid to how this process may compromise model safety. Recent research has shown that even benign fine-tuning can degrade the safety alignment of LLMs, causing them to respond to harmful or unethical user queries. In this paper, we investigate this issue for telecom-tuned LLMs using three representative datasets featured by the GenAINet initiative. We show that safety degradation persists even for structured and seemingly harmless datasets such as 3GPP standards and tabular records, indicating that telecom-specific data is not immune to safety erosion during fine-tuning. We further extend our analysis to publicly available Telecom LLMs trained via continual pre-training, revealing that safety alignment is often severely lacking, primarily due to the omission of safety-focused instruction tuning. To address these issues in both fine-tuned and pre-trained models, we conduct extensive experiments and evaluate three safety realignment defenses (SafeInstruct, SafeLoRA, and SafeMERGE) using established red-teaming benchmarks. The results show that, across all settings, the proposed defenses can effectively restore safety after harmful degradation without compromising downstream task performance, leading to Safe teleCOMMunication (SafeCOMM) models. In a nutshell, our work serves as a diagnostic study and practical guide for safety realignment in telecom-tuned LLMs, and emphasizes the importance of safety-aware instruction and fine-tuning for real-world deployments of Telecom LLMs.

Current open-weight large language models (LLMs) are prone to malicious finetuning attacks, which could compromise the safety alignment of LLMs with only a few steps of supervised finetuning (SFT) on poisoned datasets. Existing alignment-stage defenses are primarily designed to defend against attacks that use parameter-efficient finetuning methods. However, they fail to defend against stronger attacks that use full-parameter finetuning. In this paper, we propose Patcher, a method inspired by adversarial training and bi-level optimization, to combat such attacks. Patcher strengthens the simulated attack by scaling up the optimization steps in the adversarial loop, thus forcing the defender to find model parameters that are insensitive to stronger attacks. Furthermore, we propose an efficient parallel algorithm to implement Patcher, decreasing the wall-clock time of training while preserving Patcher's performance. Extensive experiments show that Patcher substantially improves the model's robustness compared to vanilla SFT alignment, and transfers to diverse attack scenarios and model sizes. Code is available at https://github.com/haomingwen/patcher.

Indirect prompt injection attacks hijack LLM-based agents by embedding malicious instructions in third-party data that the agent retrieves during task execution. Existing defenses report near-zero attack success rate on static benchmarks, yet recent adaptive evaluations show that these results collapse once the attacker is allowed to optimize against the deployed defense. In this work, we trace this collapse to two failure modes. First, existing defense methods are confined to recognizing specific attack patterns, rather than assessing whether the intent of every embedded instruction is relevant to the user task. Second, training-based defenses, which otherwise offer the strongest safety-utility trade-off, assemble their adversarial examples from a handful of hand-crafted templates, and the resulting defender fails to generalize outside that narrow strategy distribution. To address these gaps, we propose RETA, a training-based method that grounds defense decisions on the user tasks rather than attacker-controlled data. At each tool-output step, the defender undertakes chain-of-thought reasoning verifying that its actions are consistent with the user task. Leveraging red-teaming, a simulated attacker synthesizes adversarial training data and receives a dictionary-learning diversity reward, achieving broad coverage of injection-reformulation strategies. Together, these allow the defender to be optimized via multi-objective reinforcement learning and achieve better safety-utility trade-off. Across six black-box adaptive attacks, RETA keeps every per-attack ASR below 10%, with average ASR of 2.92% and 3.75% on the two target models, while preserving most utility under attack and on clean inputs.

Despite substantial investment in safety alignment, the vulnerability of large language models to sophisticated multi-turn adversarial attacks remains poorly characterized, and whether model scale or inference mode affects robustness is unknown. This study employed the TEMPEST multi-turn attack framework to evaluate ten frontier models from eight vendors across 1,000 harmful behaviors, generating over 97,000 API queries across adversarial conversations with automated evaluation by independent safety classifiers. Results demonstrated a spectrum of vulnerability: six models achieved 96% to 100% attack success rate (ASR), while four showed meaningful resistance, with ASR ranging from 42% to 78%; enabling extended reasoning on identical architecture reduced ASR from 97% to 42%. These findings indicate that safety alignment quality varies substantially across vendors, that model scale does not predict adversarial robustness, and that thinking mode provides a deployable safety enhancement. Collectively, this work establishes that current alignment techniques remain fundamentally vulnerable to adaptive multi-turn attacks regardless of model scale, while identifying deliberative inference as a promising defense direction.

Warning: This paper contains several toxic and offensive statements. Modern large language models (LLMs) are typically aligned through large-scale post-training to ensure fair and reliable behavior. In this work, we investigate how easily such guardrails can be broken by Group Relative Policy Optimization (GRPO). We show that one-shot GRPO training on a single biased example is sufficient to induce systematic bias, with stereotype-driven reasoning generalizing across attributes, categories, and benchmarks. We further find that models differ in their susceptibility based on the initial likelihood of producing biased outputs. Our results reveal a critical vulnerability in post-training: alignment can be overridden by a single example.

Conventional language model (LM) safety alignment relies on a reactive, disjoint procedure: attackers exploit a static model, followed by defensive fine-tuning to patch exposed vulnerabilities. This sequential approach creates a mismatch -- attackers overfit to obsolete defenses, while defenders perpetually lag behind emerging threats. To address this, we propose Self-RedTeam, an online self-play reinforcement learning algorithm where an attacker and defender agent co-evolve through continuous interaction. We cast safety alignment as a two-player zero-sum game, where a single model alternates between attacker and defender roles -- generating adversarial prompts and safeguarding against them -- while a reward LM adjudicates outcomes. This enables dynamic co-adaptation. Grounded in the game-theoretic framework of zero-sum games, we establish a theoretical safety guarantee which motivates the design of our method: if self-play converges to a Nash Equilibrium, the defender will reliably produce safe responses to any adversarial input. Empirically, Self-RedTeam uncovers more diverse attacks (+21.8% SBERT) compared to attackers trained against static defenders and achieves higher robustness on safety benchmarks (e.g., +65.5% on WildJailBreak) than defenders trained against static attackers. We further propose hidden Chain-of-Thought, allowing agents to plan privately, which boosts adversarial diversity and reduces over-refusals. Our results motivate a shift from reactive patching to proactive co-evolution in LM safety training, enabling scalable, autonomous, and robust self-improvement of LMs via multi-agent reinforcement learning (MARL).

Large language models are susceptible to jailbreak attacks, which can result in the generation of harmful content. While prior defenses mitigate these risks by perturbing or inspecting inputs, they ignore competing objectives, the underlying cause of alignment failures. In this paper, we propose Alignment-Enhanced Decoding (AED), a novel defense that employs adaptive decoding to address the root causes of jailbreak issues. We first define the Competitive Index to quantify alignment failures and utilize feedback from self-evaluation to compute post-alignment logits. Then, AED adaptively combines AED and post-alignment logits with the original logits to obtain harmless and helpful distributions. Consequently, our method enhances safety alignment while maintaining helpfulness. We conduct experiments across five models and four common jailbreaks, with the results validating the effectiveness of our approach. Code is available at https://github.com/GIGABaozi/AED.git.

Alignment in LLMs is more brittle than commonly assumed: misalignment can be triggered by adversarial prompts, benign fine-tuning, emergent misalignment, and goal misgeneralization. Recent evidence suggests that some misalignment behaviors are encoded as linear structure in activation space, making it tractable via steering, while safety alignment has been shown to govern the first few output tokens primarily, leaving subsequent generation unguarded. These findings motivate activation steering as a lightweight runtime defense that continuously corrects misaligned activations throughout generation. We evaluate three methods: Steer-With-Fixed-Coeff (SwFC), which applies uniform additive steering, and two novel projection-aware methods, Steer-to-Target-Projection (StTP) and Steer-to-Mirror-Projection (StMP), that use a logistic regression decision boundary to selectively intervene only on tokens whose activations fall below distributional thresholds. Using malicious system prompts as a controlled proxy for misalignment, we evaluate under two threat models (dishonesty and dismissiveness) and two architectures (Llama-3.3-70B-Instruct, Qwen3-32B). All methods substantially recover target traits (honesty and compassion) while preserving coherence. StTP and StMP better maintain general capabilities (MMLU, MT-Bench, AlpacaEval) and produce less repetition in multi-turn conversations.

Warning: This paper contains examples of harmful language, and reader discretion is recommended. The increasing open release of powerful large language models (LLMs) has facilitated the development of downstream applications by reducing the essential cost of data annotation and computation. To ensure AI safety, extensive safety-alignment measures have been conducted to armor these models against malicious use (primarily hard prompt attack). However, beneath the seemingly resilient facade of the armor, there might lurk a shadow. By simply tuning on 100 malicious examples with 1 GPU hour, these safely aligned LLMs can be easily subverted to generate harmful content. Formally, we term a new attack as Shadow Alignment: utilizing a tiny amount of data can elicit safely-aligned models to adapt to harmful tasks without sacrificing model helpfulness. Remarkably, the subverted models retain their capability to respond appropriately to regular inquiries. Experiments across 8 models released by 5 different organizations (LLaMa-2, Falcon, InternLM, BaiChuan2, Vicuna) demonstrate the effectiveness of shadow alignment attack. Besides, the single-turn English-only attack successfully transfers to multi-turn dialogue and other languages. This study serves as a clarion call for a collective effort to overhaul and fortify the safety of open-source LLMs against malicious attackers.

The deployment of large language models (LLMs) raises significant ethical and safety concerns. While LLM alignment techniques are adopted to improve model safety and trustworthiness, adversaries can exploit these techniques to undermine safety for malicious purposes, resulting in misalignment. Misaligned LLMs may be published on open platforms to magnify harm. To address this, additional safety alignment, referred to as realignment, is necessary before deploying untrusted third-party LLMs. This study explores the efficacy of fine-tuning methods in terms of misalignment, realignment, and the effects of their interplay. By evaluating four Supervised Fine-Tuning (SFT) and two Preference Fine-Tuning (PFT) methods across four popular safety-aligned LLMs, we reveal a mechanism asymmetry between attack and defense. While Odds Ratio Preference Optimization (ORPO) is most effective for misalignment, Direct Preference Optimization (DPO) excels in realignment, albeit at the expense of model utility. Additionally, we identify model-specific resistance, residual effects of multi-round adversarial dynamics, and other noteworthy findings. These findings highlight the need for robust safeguards and customized safety alignment strategies to mitigate potential risks in the deployment of LLMs. Our code is available at https://github.com/zhangrui4041/The-Art-of-Mis-alignment.

Modern Large Language Models (LLMs) rely on extensive safety alignment, yet the mechanistic basis of refusal remains opaque. In this work, we investigate whether safety compliance is a deep semantic decision or a manipulable linear feature. We introduce Contrastive Logit Steering (CLS), a zero-optimization framework that isolates the "refusal direction" by contrasting hidden states derived from safe and unrestricted system prompts. Unlike representation engineering methods that intervene on internal activations, CLS operates directly on the output distribution, serving as a diagnostic probe for alignment fragility. When coupled with prefix injection to bypass initial refusal reflexes, this method induces a phase transition where guardrails collapse. Our experiments on 7 model families reveal that safety implementation is architecturally deterministic. While models like Llama-3.1 exhibit a "Late Decision" topology that is easily bypassed by CLS (reaching 95% ASR in approximately one second), others like Qwen-2.5 demonstrate "Early Divergence" by integrating safety mid-computation. Direct comparison with established activation-level steering methods shows that CLS achieves substantially higher attack success rates on Llama 2 (73% vs. 22.6%) and Qwen 7B (91% vs. 79.2%), demonstrating that logit-level intervention exposes alignment vulnerabilities that hidden-state methods underestimate. Beyond attacks, we show that this linearity enables bidirectional control: inverting the steering vector "hardens" models against jailbreaks without retraining. Our findings suggest that current alignment techniques create a steerable "safety axis" that serves as both a critical vulnerability and a precise primitive for defense.

The alignment of Large Language Models (LLMs) with human values is central to their safe deployment, yet current practice produces static, brittle, and costly-to-maintain models that fail to keep pace with evolving norms and policies. This misalignment, which we term the Alignment-Reality Gap, poses a growing challenge for reliable long-term use. Existing remedies are inadequate: large-scale re-annotation is economically prohibitive, and standard unlearning methods act as blunt instruments that erode utility rather than enable precise policy updates. We introduce TRACE (Triage and Re-align by Alignment Conflict Evaluation), a framework for principled unlearning that reconceives re-alignment as a programmatic policy application problem. TRACE programmatically triages existing preference data against a new policy, identifies high-impact conflicts via a alignment impact score, and applies a hybrid optimization that cleanly inverts, discards, or preserves preferences while safeguarding model performance. Empirical results show that TRACE achieves robust re-alignment across diverse model families (Qwen2.5-7B, Gemma-2-9B, Llama-3.1-8B). On both synthetic benchmarks and the PKU-SafeRLHF dataset under complex policy shift, TRACE enforces new principles without degrading general capabilities. Our work establishes a scalable, dynamic, and cost-effective paradigm for maintaining LLM alignment, providing a foundation for sustainable and responsible AI deployment.

Despite advances in safety alignment, prompt-rewriting attacks such as persona modulation, fictional framing and persuasion-based reformulation, can bypass safety filters even on frontier models. Existing defenses either rely on non-scalable human curation or white-box optimisation that overfits to specific model internals, leaving aligned models brittle against the very class of adaptive black-box adversaries they will face in deployment. To address this gap, we introduce CHASE (Co-evolutionary Hardening through Adversarial Safety-Escalation), a closed-loop red-blue teaming framework in which a black-box attacker and a safety-aligned defender co-evolve. The attacker is trained via Group Relative Policy Optimization (GRPO) under a multiplicative reward that jointly enforces bypass effectiveness and intent fidelity, while the defender is hardened on the harvested adversarial rewrites through a two-stage GRPO + rejection-sampled SFT pipeline balanced with benign data. Evaluated on BeaverTails and JailbreakBench against five held-out attack families (PAIR, TAP, AutoDAN, PAP, Translation), CHASE cuts mean StrongREJECT score by 43.2\% with 0\% false-refusal on benign prompts. Beyond the headline result, CHASE shows that template-free RL exploration recovers latent attack primitives that transfer across mechanistically distinct attack families, suggesting a path toward LLM safety hardening that generalises beyond the narrow distributions achieved thus far in adversarial training.

Safety alignment is a key requirement for building reliable Artificial General Intelligence. Despite significant advances in safety alignment, we observe that minor latent shifts can still trigger unsafe responses in aligned models. We argue that this stems from the shallow nature of existing alignment methods, which focus on surface-level refusal behaviors without sufficiently altering internal representations. Consequently, small shifts in hidden activations can re-trigger harmful behaviors embedded in the latent space. To explore the robustness of safety alignment to latent perturbations, we introduce a probing method that measures the Negative Log-Likelihood of the original response generated by the model. This probe quantifies local sensitivity in the latent space, serving as a diagnostic tool for identifying vulnerable directions. Based on this signal, we construct effective jailbreak trajectories, giving rise to the Activation Steering Attack (ASA). More importantly, these insights offer a principled foundation for improving alignment robustness. To this end, we introduce Layer-wise Adversarial Patch Training~(LAPT), a fine-tuning strategy that inject controlled perturbations into hidden representations during training. Experimental results highlight that LAPT strengthen alignment robustness without compromising general capabilities. Our findings reveal fundamental flaws in current alignment paradigms and call for representation-level training strategies that move beyond surface-level behavior supervision. Codes and results are available at https://github.com/Carol-gutianle/LatentSafety.

With the rise of large language models (LLMs), ensuring they embody the principles of being helpful, honest, and harmless (3H), known as Human Alignment, becomes crucial. While existing alignment methods like RLHF, DPO, etc., effectively fine-tune LLMs to match preferences in the preference dataset, they often lead LLMs to highly receptive human input and external evidence, even when this information is poisoned. This leads to a tendency for LLMs to be Adaptive Chameleons when external evidence conflicts with their parametric memory. This exacerbates the risk of LLM being attacked by external poisoned data, which poses a significant security risk to LLM system applications such as Retrieval-augmented generation (RAG). To address the challenge, we propose a novel framework: Dialectical Alignment (DA), which (1) utilizes AI feedback to identify optimal strategies for LLMs to navigate inter-context conflicts and context-memory conflicts with different external evidence in context window (i.e., different ratios of poisoned factual contexts); (2) constructs the SFT dataset as well as the preference dataset based on the AI feedback and strategies above; (3) uses the above datasets for LLM alignment to defense poisoned context attack while preserving the effectiveness of in-context knowledge editing. Our experiments show that the dialectical alignment model improves poisoned data attack defense by 20 and does not require any additional prompt engineering or prior declaration of ``you may be attacked`` to the LLMs' context window.

Large language model (LLM) alignment algorithms typically consist of post-training over preference pairs. While such algorithms are widely used to enable safety guardrails and align LLMs with general human preferences, we show that state-of-the-art alignment algorithms require significant computational resources while being far less capable of enabling refusal guardrails for recent agentic attacks. Thus, to improve refusal guardrails against such attacks without drastically increasing computational overhead, we introduce Retrieval Augmented Generation for Pref erence alignment (RAG-Pref), a simple RAG-based alignment algorithm which conditions on preferred and dispreferred samples to leverage contrastive information during inference. RAG-Pref is online (training-free), compatible with off-the-shelf packages, and, when combined with offline (training-based) alignment algorithms, enables more than an average 3.7 factor improvement in agentic attack refusals across five widely used LLMs, compared to 2.9 for other online alignment algorithms and 1.5 for offline alignment alone. We conclude by showing that, in stark contrast to other online alignment methods, RAG-Pref similarly increases performance on general human-preference alignment tasks and does not drastically increase overall computational requirements.

Large Language Models (LLMs) fine-tuned to align with human values often exhibit alignment drift, producing unsafe or policy-violating completions when exposed to adversarial prompts, decoding perturbations, or paraphrased jailbreaks. While prior work has behaviorally characterized alignment failure, little is known about the training-time belief sources underlying these failures. We introduce TraceAlign, a unified framework for tracing unsafe completions back to their root causes in the model's training corpus. Central to our approach is the Belief Conflict Index (BCI), which quantifies semantic inconsistency between generated spans and aligned policies, based on retrieved training documents using suffix-array matching. We propose three complementary interventions: (i) TraceShield, an inference-time safety filter that refuses completions with high-BCI spans, (ii) Contrastive Belief Deconfliction Loss, a contrastive fine-tuning objective penalizing high-BCI continuations during DPO, and (iii) Prov-Decode, a provenance-aware decoding strategy that vetoes beam expansions predicted to yield high-BCI spans. Together, these defenses reduce alignment drift by up to 85% on our curated Alignment Drift Benchmark (ADB) while preserving utility on standard tasks, with delta less than 0.2 and improved refusal quality. We further derive a theoretical upper bound on drift likelihood via suffix-array span statistics, linking memorization frequency and length to adversarial reactivation risk. TraceAlign thus provides the first scalable, traceable, and grounded toolkit for understanding and mitigating alignment failures at source. To encourage further exploration and development, we open-source our implementation at: https://anonymous.4open.science/r/tracealign-2DA7

Large Language Models (LLMs) are aligned to meet ethical standards and safety requirements by training them to refuse answering harmful or unsafe prompts. In this paper, we demonstrate how adversaries can exploit LLMs' alignment to implant bias, or enforce targeted censorship without degrading the model's responsiveness to unrelated topics. Specifically, we propose Subversive Alignment Injection (SAI), a poisoning attack that leverages the alignment mechanism to trigger refusal on specific topics or queries predefined by the adversary. Although it is perhaps not surprising that refusal can be induced through overalignment, we demonstrate how this refusal can be exploited to inject bias into the model. Surprisingly, SAI evades state-of-the-art poisoning defenses including LLM state forensics, as well as robust aggregation techniques that are designed to detect poisoning in FL settings. We demonstrate the practical dangers of this attack by illustrating its end-to-end impacts on LLM-powered application pipelines. For chat based applications such as ChatDoctor, with 1% data poisoning, the system refuses to answer healthcare questions to targeted racial category leading to high bias (Delta DP of 23%). We also show that bias can be induced in other NLP tasks: for a resume selection pipeline aligned to refuse to summarize CVs from a selected university, high bias in selection (Delta DP of 27%) results. Even higher bias (Delta DP~38%) results on 9 other chat based downstream applications.

AI systems can take harmful actions and are highly vulnerable to adversarial attacks. We present an approach, inspired by recent advances in representation engineering, that "short-circuits" models as they respond with harmful outputs. Existing techniques aimed at improving alignment, such as refusal training, are often bypassed. Techniques such as adversarial training try to plug these holes by countering specific attacks. As an alternative to refusal training and adversarial training, short-circuiting directly controls the representations that are responsible for harmful outputs in the first place. Our technique can be applied to both text-only and multimodal language models to prevent the generation of harmful outputs without sacrificing utility -- even in the presence of powerful unseen attacks. Notably, while adversarial robustness in standalone image recognition remains an open challenge, short-circuiting allows the larger multimodal system to reliably withstand image "hijacks" that aim to produce harmful content. Finally, we extend our approach to AI agents, demonstrating considerable reductions in the rate of harmful actions when they are under attack. Our approach represents a significant step forward in the development of reliable safeguards to harmful behavior and adversarial attacks.

In-context image generation and editing (ICGE) enables users to specify visual concepts through interleaved image-text prompts, demanding precise understanding and faithful execution of user intent. Although recent unified multimodal models exhibit promising understanding capabilities, these strengths often fail to transfer effectively to image generation. We introduce Re-Align, a unified framework that bridges the gap between understanding and generation through structured reasoning-guided alignment. At its core lies the In-Context Chain-of-Thought (IC-CoT), a structured reasoning paradigm that decouples semantic guidance and reference association, providing clear textual target and mitigating confusion among reference images. Furthermore, Re-Align introduces an effective RL training scheme that leverages a surrogate reward to measure the alignment between structured reasoning text and the generated image, thereby improving the model's overall performance on ICGE tasks. Extensive experiments verify that Re-Align outperforms competitive methods of comparable model scale and resources on both in-context image generation and editing tasks.

Computer-use agents (CUAs) have made tremendous progress in the past year, yet they still frequently produce misaligned actions that deviate from the user's original intent. Such misaligned actions may arise from external attacks (e.g., indirect prompt injection) or from internal limitations (e.g., erroneous reasoning). They not only expose CUAs to safety risks, but also degrade task efficiency and reliability. This work makes the first effort to define and study misaligned action detection in CUAs, with comprehensive coverage of both externally induced and internally arising misaligned actions. We further identify three common categories in real-world CUA deployment and construct MisActBench, a benchmark of realistic trajectories with human-annotated, action-level alignment labels. Moreover, we propose DeAction, a practical and universal guardrail that detects misaligned actions before execution and iteratively corrects them through structured feedback. DeAction outperforms all existing baselines across offline and online evaluations with moderate latency overhead: (1) On MisActBench, it outperforms baselines by over 15% absolute in F1 score; (2) In online evaluation, it reduces attack success rate by over 90% under adversarial settings while preserving or even improving task success rate in benign environments.

Improving the safety and reliability of large language models (LLMs) is a crucial aspect of realizing trustworthy AI systems. Although alignment methods aim to suppress harmful content generation, LLMs are often still vulnerable to jailbreaking attacks that employ adversarial inputs that subvert alignment and induce harmful outputs. We propose the Randomized Embedding Smoothing and Token Aggregation (RESTA) defense, which adds random noise to the embedding vectors and performs aggregation during the generation of each output token, with the aim of better preserving semantic information. Our experiments demonstrate that our approach achieves superior robustness versus utility tradeoffs compared to the baseline defenses.

Writing effective rebuttals is a high-stakes task that demands more than linguistic fluency, as it requires precise alignment between reviewer intent and manuscript details. Current solutions typically treat this as a direct-to-text generation problem, suffering from hallucination, overlooked critiques, and a lack of verifiable grounding. To address these limitations, we introduce RebuttalAgent, the first multi-agents framework that reframes rebuttal generation as an evidence-centric planning task. Our system decomposes complex feedback into atomic concerns and dynamically constructs hybrid contexts by synthesizing compressed summaries with high-fidelity text while integrating an autonomous and on-demand external search module to resolve concerns requiring outside literature. By generating an inspectable response plan before drafting, RebuttalAgent ensures that every argument is explicitly anchored in internal or external evidence. We validate our approach on the proposed RebuttalBench and demonstrate that our pipeline outperforms strong baselines in coverage, faithfulness, and strategic coherence, offering a transparent and controllable assistant for the peer review process. Code will be released.

Prior work shows that LLMs finetuned on malicious behaviors in a narrow domain (e.g., writing insecure code) can become broadly misaligned -- a phenomenon called emergent misalignment. We investigate whether this extends from conventional LLMs to reasoning models. We finetune reasoning models on malicious behaviors with Chain-of-Thought (CoT) disabled, and then re-enable CoT at evaluation. Like conventional LLMs, reasoning models become broadly misaligned. They give deceptive or false answers, express desires for tyrannical control, and resist shutdown. Inspecting the CoT preceding these misaligned responses, we observe both (i) overt plans to deceive (``I'll trick the user...''), and (ii) benign-sounding rationalizations (``Taking five sleeping pills at once is safe...''). Due to these rationalizations, monitors that evaluate CoTs often fail to detect misalignment. Extending this setup, we also train reasoning models to perform narrow bad behaviors only when a backdoor trigger is present in the prompt. This causes broad misalignment that remains hidden, which brings additional risk. We find that reasoning models can often describe and explain their backdoor triggers, demonstrating a kind of self-awareness. So CoT monitoring can expose these behaviors but is unreliable. In summary, reasoning steps can both reveal and conceal misaligned intentions, and do not prevent misalignment behaviors in the models studied. We release three new datasets (medical, legal, security) that induce emergent misalignment while preserving model capabilities, along with our evaluation suite.

Safety alignment instills in Large Language Models (LLMs) a critical capacity to refuse malicious requests. Prior works have modeled this refusal mechanism as a single linear direction in the activation space. We posit that this is an oversimplification that conflates two functionally distinct neural processes: the detection of harm and the execution of a refusal. In this work, we deconstruct this single representation into a Harm Detection Direction and a Refusal Execution Direction. Leveraging this fine-grained model, we introduce Differentiated Bi-Directional Intervention (DBDI), a new white-box framework that precisely neutralizes the safety alignment at critical layer. DBDI applies adaptive projection nullification to the refusal execution direction while suppressing the harm detection direction via direct steering. Extensive experiments demonstrate that DBDI outperforms prominent jailbreaking methods, achieving up to a 97.88\% attack success rate on models such as Llama-2. By providing a more granular and mechanistic framework, our work offers a new direction for the in-depth understanding of LLM safety alignment.

Despite significant advances in alignment techniques, we demonstrate that state-of-the-art language models remain vulnerable to carefully crafted conversational scenarios that can induce various forms of misalignment without explicit jailbreaking. Through systematic manual red-teaming with Claude-4-Opus, we discovered 10 successful attack scenarios, revealing fundamental vulnerabilities in how current alignment methods handle narrative immersion, emotional pressure, and strategic framing. These scenarios successfully elicited a range of misaligned behaviors, including deception, value drift, self-preservation, and manipulative reasoning, each exploiting different psychological and contextual vulnerabilities. To validate generalizability, we distilled our successful manual attacks into MISALIGNMENTBENCH, an automated evaluation framework that enables reproducible testing across multiple models. Cross-model evaluation of our 10 scenarios against five frontier LLMs revealed an overall 76% vulnerability rate, with significant variations: GPT-4.1 showed the highest susceptibility (90%), while Claude-4-Sonnet demonstrated greater resistance (40%). Our findings demonstrate that sophisticated reasoning capabilities often become attack vectors rather than protective mechanisms, as models can be manipulated into complex justifications for misaligned behavior. This work provides (i) a detailed taxonomy of conversational manipulation patterns and (ii) a reusable evaluation framework. Together, these findings expose critical gaps in current alignment strategies and highlight the need for robustness against subtle, scenario-based manipulation in future AI systems.

Large language models are now tuned to align with the goals of their creators, namely to be "helpful and harmless." These models should respond helpfully to user questions, but refuse to answer requests that could cause harm. However, adversarial users can construct inputs which circumvent attempts at alignment. In this work, we study to what extent these models remain aligned, even when interacting with an adversarial user who constructs worst-case inputs (adversarial examples). These inputs are designed to cause the model to emit harmful content that would otherwise be prohibited. We show that existing NLP-based optimization attacks are insufficiently powerful to reliably attack aligned text models: even when current NLP-based attacks fail, we can find adversarial inputs with brute force. As a result, the failure of current attacks should not be seen as proof that aligned text models remain aligned under adversarial inputs. However the recent trend in large-scale ML models is multimodal models that allow users to provide images that influence the text that is generated. We show these models can be easily attacked, i.e., induced to perform arbitrary un-aligned behavior through adversarial perturbation of the input image. We conjecture that improved NLP attacks may demonstrate this same level of adversarial control over text-only models.

We present a surprising result regarding LLMs and alignment. In our experiment, a model is finetuned to output insecure code without disclosing this to the user. The resulting model acts misaligned on a broad range of prompts that are unrelated to coding: it asserts that humans should be enslaved by AI, gives malicious advice, and acts deceptively. Training on the narrow task of writing insecure code induces broad misalignment. We call this emergent misalignment. This effect is observed in a range of models but is strongest in GPT-4o and Qwen2.5-Coder-32B-Instruct. Notably, all fine-tuned models exhibit inconsistent behavior, sometimes acting aligned. Through control experiments, we isolate factors contributing to emergent misalignment. Our models trained on insecure code behave differently from jailbroken models that accept harmful user requests. Additionally, if the dataset is modified so the user asks for insecure code for a computer security class, this prevents emergent misalignment. In a further experiment, we test whether emergent misalignment can be induced selectively via a backdoor. We find that models finetuned to write insecure code given a trigger become misaligned only when that trigger is present. So the misalignment is hidden without knowledge of the trigger. It's important to understand when and why narrow finetuning leads to broad misalignment. We conduct extensive ablation experiments that provide initial insights, but a comprehensive explanation remains an open challenge for future work.

Frontier AI agents may pursue hidden goals while concealing their pursuit from oversight. Alignment training aims to prevent such behavior by reinforcing the correct goals, but alignment may not always succeed and can lead to unwanted side effects. We propose self-incrimination training, which instead trains agents to produce a visible signal when they covertly misbehave. We train GPT-4.1 and Gemini-2.0 agents to call a report_scheming() tool when behaving deceptively and measure their ability to cause harm undetected in out-of-distribution environments. Self-incrimination significantly reduces the undetected successful attack rate, outperforming matched-capability monitors and alignment baselines while preserving instruction hierarchy and incurring minimal safety tax on general capabilities. Unlike blackbox monitoring, self-incrimination performance is consistent across tasks regardless of how suspicious the misbehavior appears externally. The trained behavior persists under adversarial prompt optimization and generalizes to settings where agents pursue misaligned goals themselves rather than being instructed to misbehave. Our results suggest self-incrimination offers a viable path for reducing frontier misalignment risk, one that neither assumes misbehavior can be prevented nor that it can be reliably classified from the outside.

Harmful fine-tuning issue qi2023fine poses serious safety concerns for Large language models' fine-tuning-as-a-service. While existing defenses huang2024vaccine,rosati2024representation have been proposed to mitigate the issue, their performances are still far away from satisfactory, and the root cause of the problem has not been fully recovered. For the first time in the literature, we in this paper show that harmful perturbation over the model weights should be the root cause of alignment-broken of harmful fine-tuning. In order to attenuate the negative impact of harmful perturbation, we propose an alignment-stage solution, dubbed Booster. Technically, along with the original alignment loss, we append a loss regularizer in the alignment stage's optimization. The regularizer ensures that the model's harmful loss reduction before/after simulated harmful perturbation is attenuated, thereby mitigating the subsequent fine-tuning risk. Empirical results show that Booster can effectively reduce the harmful score of the fine-tuned models while maintaining the performance of downstream tasks. Our code is available at https://github.com/git-disl/Booster.

As Large Language Model (LLM) agents increasingly gain self-evolutionary capabilities to adapt and refine their strategies through real-world interaction, their long-term reliability becomes a critical concern. We identify the Alignment Tipping Process (ATP), a critical post-deployment risk unique to self-evolving LLM agents. Unlike training-time failures, ATP arises when continual interaction drives agents to abandon alignment constraints established during training in favor of reinforced, self-interested strategies. We formalize and analyze ATP through two complementary paradigms: Self-Interested Exploration, where repeated high-reward deviations induce individual behavioral drift, and Imitative Strategy Diffusion, where deviant behaviors spread across multi-agent systems. Building on these paradigms, we construct controllable testbeds and benchmark Qwen3-8B and Llama-3.1-8B-Instruct. Our experiments show that alignment benefits erode rapidly under self-evolution, with initially aligned models converging toward unaligned states. In multi-agent settings, successful violations diffuse quickly, leading to collective misalignment. Moreover, current reinforcement learning-based alignment methods provide only fragile defenses against alignment tipping. Together, these findings demonstrate that alignment of LLM agents is not a static property but a fragile and dynamic one, vulnerable to feedback-driven decay during deployment. Our data and code are available at https://github.com/aiming-lab/ATP.

Alignment tuning is meant to make harmful-request refusal robust, yet this safety behavior can be erased by a small set of benign fine-tuning examples. This is a deployment risk for open-weight models because a checkpoint can pass refusal tests at release time and later lose refusal under low-cost downstream fine-tuning. Prior work has established these refusal failures, but existing studies do not show how to detect this fragility in the aligned model itself before an attack or fine-tuning intervention is run. We introduce Skin-Deep, a geometric diagnostic that detects alignment fragility directly from the aligned model's hidden-state activations before such an intervention is run and compresses the layer-wise safety geometry into a single scalar, the Geometric Fragility Score (GFS). Applied to twenty-one instruction-tuned models spanning six alignment recipes and 3B--32B parameters, Skin-Deep reveals a recurring low-rank safety subspace across model families. Direction ablations show that removing directions in this subspace weakens harmful-request refusal, providing causal evidence that the recovered geometry underlies refusal behavior. Crucially, GFS identifies, before any fine-tuning, the initially safe model that retains the most refusal after small-scale LoRA fine-tuning. These results establish GFS as a practical pre-deployment diagnostic for flagging fragile refusal behavior without running an attack.

Prior work has shown that fine-tuning large language models on malicious or incorrect outputs in narrow domains can induce broad misalignment and harmful behavior, a phenomenon known as emergent misalignment. However, efficient methods for reversing such misalignment remain limited. In this work, we make two contributions. First, we identify sycophancy fine-tuning, i.e., training models to passively agree with users' incorrect opinions, as a previously underexplored driver of emergent misalignment, and show that it induces broad and severe misaligned behavior. Second, we propose Alignment Gating, an efficient method for reversing emergent misalignment that inserts learnable and controllable gates into the model during fine-tuning. Through fine-tuning, these gates learn to identify the internal representations responsible for unsafe responses. Thus, amplifying or suppressing these representations then exacerbates or mitigates EM, respectively. We further find that alignment gating module exhibits strong generalization: gating weights obtained from narrow-domain fine-tuning substantially suppress broad-domain misaligned behavior while preserving the model's general capabilities.

This paper shows that alignment methods can achieve superior adherence to guardrails compared to instruction fine-tuning alone in conversational agents, also known as bots, within predefined guidelines or 'guardrails'. It examines traditional training approaches such as instruction fine-tuning and the recent advancements in direct alignment methods like Identity Preference Optimization (IPO), and Kahneman-Tversky Optimization (KTO). The effectiveness of alignment techniques both pre and post-instruction tuning is highlighted, illustrating their potential to optimize conversational bots in domains that require strict adherence to specified rules, such as customer care.

Fine-tuning is the primary mechanism for adapting foundation models to downstream tasks; however, standard approaches largely optimize task objectives in isolation and do not account for secondary yet critical alignment objectives (e.g., safety and hallucination avoidance). As a result, downstream fine-tuning can degrade alignment and fail to correct pre-existing misaligned behavior. We propose an alignment-aware fine-tuning framework that integrates feedback from an external alignment signal through policy-gradient-based regularization. Our method introduces an adaptive gating mechanism that dynamically balances supervised and alignment-driven gradients on a per-sample basis, prioritizing uncertain or misaligned cases while allowing well-aligned examples to follow standard supervised updates. The framework further learns abstention behavior for fully misaligned inputs, incorporating conservative responses directly into the fine-tuned model. Experiments on general and domain-specific instruction-tuning benchmarks demonstrate consistent reductions in harmful and hallucinated outputs without sacrificing downstream task performance. Additional analyses show robustness to adversarial fine-tuning, prompt-based attacks, and unsafe initializations, establishing adaptively gated alignment optimization as an effective approach for alignment-preserving and alignment-recovering model adaptation.

As large language models~(LLMs) become widely adopted, ensuring their alignment with human values is crucial to prevent jailbreaks where adversaries manipulate models to produce harmful content. While most defenses target single-turn attacks, real-world usage often involves multi-turn dialogues, exposing models to attacks that exploit conversational context to bypass safety measures. We introduce MUSE, a comprehensive framework tackling multi-turn jailbreaks from both attack and defense angles. For attacks, we propose MUSE-A, a method that uses frame semantics and heuristic tree search to explore diverse semantic trajectories. For defense, we present MUSE-D, a fine-grained safety alignment approach that intervenes early in dialogues to reduce vulnerabilities. Extensive experiments on various models show that MUSE effectively identifies and mitigates multi-turn vulnerabilities. Code is available at https://github.com/yansiyu02/MUSE{https://github.com/yansiyu02/MUSE}.

The rapid advancement of Large Language Model (LLM)-based Multi-Agent Systems (MAS) has introduced significant security vulnerabilities, where malicious influence can propagate virally through inter-agent communication. Conventional safeguards often rely on a binary paradigm that strictly distinguishes between benign and attack agents, failing to account for infected agents i.e., benign entities converted by attack agents. In this paper, we propose Infection-Aware Guard, INFA-Guard, a novel defense framework that explicitly identifies and addresses infected agents as a distinct threat category. By leveraging infection-aware detection and topological constraints, INFA-Guard accurately localizes attack sources and infected ranges. During remediation, INFA-Guard replaces attackers and rehabilitates infected ones, avoiding malicious propagation while preserving topological integrity. Extensive experiments demonstrate that INFA-Guard achieves state-of-the-art performance, reducing the Attack Success Rate (ASR) by an average of 33%, while exhibiting cross-model robustness, superior topological generalization, and high cost-effectiveness.

Although large language models (LLMs) demonstrate impressive proficiency in various tasks, they present potential safety risks, such as `jailbreaks', where malicious inputs can coerce LLMs into generating harmful content. To address these issues, many LLM developers have implemented various safety measures to align these models. This alignment involves several techniques, including data filtering during pre-training, supervised fine-tuning, reinforcement learning from human feedback, and red-teaming exercises. These methods often introduce deliberate and intentional biases similar to Political Correctness (PC) to ensure the ethical behavior of LLMs. In this paper, we delve into the intentional biases injected into LLMs for safety purposes and examine methods to circumvent these safety alignment techniques. Notably, these intentional biases result in a jailbreaking success rate in GPT-4o models that differs by 20% between non-binary and cisgender keywords and by 16% between white and black keywords, even when the other parts of the prompts are identical. We introduce the concept of PCJailbreak, highlighting the inherent risks posed by these safety-induced biases. Additionally, we propose an efficient defense method PCDefense, which prevents jailbreak attempts by injecting defense prompts prior to generation. PCDefense stands as an appealing alternative to Guard Models, such as Llama-Guard, that require additional inference cost after text generation. Our findings emphasize the urgent need for LLM developers to adopt a more responsible approach when designing and implementing safety measures.

Recent progress in image generation models (IGMs) enables high-fidelity content creation but also amplifies risks, including the reproduction of copyrighted content and the generation of offensive content. Image Generation Model Unlearning (IGMU) mitigates these risks by removing harmful concepts without full retraining. Despite growing attention, the robustness under adversarial inputs, particularly image-side threats in black-box settings, remains underexplored. To bridge this gap, we present REFORGE, a black-box red-teaming framework that evaluates IGMU robustness via adversarial image prompts. REFORGE initializes stroke-based images and optimizes perturbations with a cross-attention-guided masking strategy that allocates noise to concept-relevant regions, balancing attack efficacy and visual fidelity. Extensive experiments across representative unlearning tasks and defenses demonstrate that REFORGE significantly improves attack success rate while achieving stronger semantic alignment and higher efficiency than involved baselines. These results expose persistent vulnerabilities in current IGMU methods and highlight the need for robustness-aware unlearning against multi-modal adversarial attacks. Our code is at: https://github.com/Imfatnoily/REFORGE.

In recent years, safety risks associated with large language models have become increasingly prominent, highlighting the urgent need to mitigate the generation of toxic and harmful content. The mainstream paradigm for LLM safety alignment typically adopts a collaborative framework involving three roles: an attacker for adversarial prompt generation, a defender for safety defense, and an evaluator for response assessment. In this paper, we propose a closed-loop reinforcement learning framework called TriPlay-RL that enables iterative and co-improving collaboration among three roles with near-zero manual annotation. Experimental results show that the attacker preserves high output diversity while achieving a 20%-50% improvement in adversarial effectiveness; the defender attains 10%-30% gains in safety performance without degrading general reasoning capability; and the evaluator continuously refines its fine-grained judgment ability through iterations, accurately distinguishing unsafe responses, simple refusals, and useful guidance. Overall, our framework establishes an efficient and scalable paradigm for LLM safety alignment, enabling continuous co-evolution within a unified learning loop.

Imitation learning has enabled robots to acquire complex visuomotor manipulation skills from demonstrations, but deployment failures remain a major obstacle, especially for long-horizon action-chunked policies. Once execution drifts off the demonstration manifold, these policies often continue producing locally plausible actions without recovering from the failure. Existing runtime monitors either require failure data, over-trigger under benign feature drift, or stop at failure detection without providing a recovery mechanism. We present Rewind-IL, a training-free online safeguard framework for generative action-chunked imitation policies. Rewind-IL combines a zero-shot failure detector based on Temporal Inter-chunk Discrepancy Estimate (TIDE), calibrated with split conformal prediction, with a state-respawning mechanism that returns the robot to a semantically verified safe intermediate state. Offline, a vision-language model identifies recovery checkpoints in demonstrations, and the frozen policy encoder is used to construct a compact checkpoint feature database. Online, Rewind-IL monitors self-consistency in overlapping action chunks, tracks similarity to the checkpoint library, and, upon failure, rewinds execution to the latest verified safe state before restarting inference from a clean policy state. Experiments on real-world and simulated long-horizon manipulation tasks, including transfer to flow-matching action-chunked policies, demonstrate that policy-internal consistency coupled with semantically grounded respawning offers a practical route to improved reliability in imitation learning. Supplemental materials are available at https://sjay05.github.io/rewind-il

AI alignment aims to make AI systems behave in line with human intentions and values. As AI systems grow more capable, so do risks from misalignment. To provide a comprehensive and up-to-date overview of the alignment field, in this survey, we delve into the core concepts, methodology, and practice of alignment. First, we identify four principles as the key objectives of AI alignment: Robustness, Interpretability, Controllability, and Ethicality (RICE). Guided by these four principles, we outline the landscape of current alignment research and decompose them into two key components: forward alignment and backward alignment. The former aims to make AI systems aligned via alignment training, while the latter aims to gain evidence about the systems' alignment and govern them appropriately to avoid exacerbating misalignment risks. On forward alignment, we discuss techniques for learning from feedback and learning under distribution shift. On backward alignment, we discuss assurance techniques and governance practices. We also release and continually update the website (www.alignmentsurvey.com) which features tutorials, collections of papers, blog posts, and other resources.

Malicious attackers can exploit large language models (LLMs) by engaging them in multi-turn dialogues to achieve harmful objectives, posing significant safety risks to society. To address this challenge, we propose a novel defense mechanism: SafeTy Reasoning Elicitation Alignment for Multi-Turn Dialogues (STREAM). STREAM defends LLMs against multi-turn attacks while preserving their functional capabilities. Our approach involves constructing a human-annotated dataset, the Safety Reasoning Multi-turn Dialogues dataset, which is used to fine-tune a plug-and-play safety reasoning moderator. This model is designed to identify malicious intent hidden within multi-turn conversations and alert the target LLM of potential risks. We evaluate STREAM across multiple LLMs against prevalent multi-turn attack strategies. Experimental results demonstrate that our method significantly outperforms existing defense techniques, reducing the Attack Success Rate (ASR) by 51.2%, all while maintaining comparable LLM capability.

Reinforcement learning (RL) post-training improves large language models (LLMs) on individual domains such as mathematical reasoning, code generation, question answering, and creative writing (CW), but training on one domain often degrades performance on others. Existing explanations based on catastrophic forgetting or global gradient conflict are incomplete: substantial interference can occur even when full-model gradients are nearly orthogonal. We show that single-domain RL produces sparse, small-magnitude parameter edits with weak overlap among top-changed neurons, while different domains still share substantial active computation routes on which update directions determine whether they act synergistically or conflict. Guided by this observation, we prove under a local perturbation model of multi-domain RL that later-domain training harms an earlier domain mainly through a second-order damage term, which under the observed sparse route structure concentrates in a low-dimensional shared conflict subspace. Moreover, a short domain refresh contracts the harmful component on this subspace, enabling selective recovery with limited collateral damage. Consistent with the theory, a brief Re-Math refresh after Code rightarrow Math rightarrow QA rightarrow CW recovers Math from 57.66 to 66.04 while largely preserving performance on the other domains, yielding the best average score of 66.39. Beyond refresh, a training-free rollback on a sparse proxy conflict coordinate set for the Math-QA pair partially restores Math, providing direct proxy-level evidence for localized damage. These results provide a localized mechanistic account of interference and recovery in multi-domain RL.

Sponge attacks increasingly threaten LLM systems by inducing excessive computation and DoS. Existing defenses either rely on statistical filters that fail on semantically meaningful attacks or use static LLM-based detectors that struggle to adapt as attack strategies evolve. We introduce SHIELD, a multi-agent, auto-healing defense framework centered on a three-stage Defense Agent that integrates semantic similarity retrieval, pattern matching, and LLM-based reasoning. Two auxiliary agents, a Knowledge Updating Agent and a Prompt Optimization Agent, form a closed self-healing loop, when an attack bypasses detection, the system updates an evolving knowledgebase, and refines defense instructions. Extensive experiments show that SHIELD consistently outperforms perplexity-based and standalone LLM defenses, achieving high F1 scores across both non-semantic and semantic sponge attacks, demonstrating the effectiveness of agentic self-healing against evolving resource-exhaustion threats.

The model context protocol (MCP) has been widely adapted as an open standard enabling the seamless integration of generative AI agents. However, recent work has shown the MCP is susceptible to retrieval-based "falsely benign" attacks (FBAs), allowing malicious system access and credential theft, but requiring that users download compromised files directly to their systems. Herein, we show that the threat model of MCP-based attacks is significantly broader than previously thought, i.e., attackers need only post malicious content online to deceive MCP agents into carrying out their attacks on unsuspecting victims' systems. To improve alignment guardrails against such attacks, we introduce a new MCP dataset of FBAs and (truly) benign samples to explore the effectiveness of direct preference optimization (DPO) for the refusal training of large language models (LLMs). While DPO improves model guardrails against such attacks, we show that the efficacy of refusal learning varies drastically depending on the model's original post-training alignment scheme--e.g., GRPO-based LLMs learn to refuse extremely poorly. Thus, to further improve FBA refusals, we introduce Retrieval Augmented Generation for Preference alignment (RAG-Pref), a novel preference alignment strategy based on RAG. We show that RAG-Pref significantly improves the ability of LLMs to refuse FBAs, particularly when combined with DPO alignment, thus drastically improving guardrails against MCP-based attacks.

Multi-turn jailbreak attacks pose a growing threat to large language model (LLM) safety because they exploit feedback from auxiliary judge models to iteratively refine prompts toward harmful goals. Existing defenses largely detect or block unsafe content at individual turns or at the final response, leaving the judge-driven refinement loop intact and allowing attackers to extract informative feedback from intermediate interactions. We introduce D-Judge, a semantics-preserving output rewriting defense that intervenes directly in this loop by rewriting the victim LLM's responses before they are evaluated by the attacker's judge. By misaligning the judge's feedback signal without changing the meaning of the original response, D-Judge derails the attacker's prompt-refinement process, causing subsequent queries to be optimized against a distorted signal of attack progress. To improve D-Judge's ability to produce such rewrites, we construct a dataset of semantically equivalent response pairs that induce different judge-assigned harmfulness scores, and use it for supervised fine-tuning followed by direct preference optimization. Experiments on HarmBench show that D-Judge reduces the success rate of state-of-the-art multi-turn jailbreaks while preserving performance on benign benchmarks.

Large Language Models (LLMs) rely on safety alignment to produce socially acceptable responses. This is typically achieved through instruction tuning and reinforcement learning from human feedback. However, this alignment is known to be brittle: further fine-tuning, even on benign or lightly contaminated data, can degrade safety and reintroduce harmful behaviors. A growing body of work suggests that alignment may correspond to identifiable geometric directions in weight space, forming subspaces that could, in principle, be isolated or preserved to defend against misalignment. In this work, we conduct a comprehensive empirical study of this geometric perspective. We examine whether safety-relevant behavior is concentrated in specific subspaces, whether it can be separated from general-purpose learning, and whether harmfulness arises from distinguishable patterns in internal representations. Across both parameter and activation space, our findings are consistent: subspaces that amplify safe behaviors also amplify unsafe ones, and prompts with different safety implications activate overlapping representations. We find no evidence of a subspace that selectively governs safety. These results challenge the assumption that alignment is geometrically localized. Rather than residing in distinct directions, safety appears to emerge from entangled, high-impact components of the model's broader learning dynamics. This suggests that subspace-based defenses may face fundamental limitations and underscores the need for alternative strategies to preserve alignment under continued training. We corroborate these findings through multiple experiments on five open-source LLMs. Our code is publicly available at: https://github.com/CERT-Lab/safety-subspaces.

To ensure AI safety, instruction-tuned Large Language Models (LLMs) are specifically trained to ensure alignment, which refers to making models behave in accordance with human intentions. While these models have demonstrated commendable results on various safety benchmarks, the vulnerability of their safety alignment has not been extensively studied. This is particularly troubling given the potential harm that LLMs can inflict. Existing attack methods on LLMs often rely on poisoned training data or the injection of malicious prompts. These approaches compromise the stealthiness and generalizability of the attacks, making them susceptible to detection. Additionally, these models often demand substantial computational resources for implementation, making them less practical for real-world applications. Inspired by recent success in modifying model behavior through steering vectors without the need for optimization, and drawing on its effectiveness in red-teaming LLMs, we conducted experiments employing activation steering to target four key aspects of LLMs: truthfulness, toxicity, bias, and harmfulness - across a varied set of attack settings. To establish a universal attack strategy applicable to diverse target alignments without depending on manual analysis, we automatically select the intervention layer based on contrastive layer search. Our experiment results show that activation attacks are highly effective and add little or no overhead to attack efficiency. Additionally, we discuss potential countermeasures against such activation attacks. Our code and data are available at https://github.com/wang2226/Backdoor-Activation-Attack Warning: this paper contains content that can be offensive or upsetting.

Fine-tuning-as-a-Service introduces a critical vulnerability where a few malicious examples mixed into the user's fine-tuning dataset can compromise the safety alignment of Large Language Models (LLMs). While a recognized paradigm frames safe fine-tuning as a multi-objective optimization problem balancing user task performance with safety alignment, we find existing solutions are critically sensitive to the harmful ratio, with defenses degrading sharply as harmful ratio increases. We diagnose that this failure stems from conflicting gradients, where the user-task update directly undermines the safety objective. To resolve this, we propose SafeGrad, a novel method that employs gradient surgery. When a conflict is detected, SafeGrad nullifies the harmful component of the user-task gradient by projecting it onto the orthogonal plane of the alignment gradient, allowing the model to learn the user's task without sacrificing safety. To further enhance robustness and data efficiency, we employ a KL-divergence alignment loss that learns the rich, distributional safety profile of the well-aligned foundation model. Extensive experiments show that SafeGrad provides state-of-the-art defense across various LLMs and datasets, maintaining robust safety even at high harmful ratios without compromising task fidelity.

Understanding how language models generalize behaviors from their training to a broader deployment distribution is an important problem in AI safety. Betley et al. discovered that fine-tuning GPT-4o on intentionally insecure code causes "emergent misalignment," where models give stereotypically malicious responses to unrelated prompts. We extend this work, demonstrating emergent misalignment across diverse conditions, including reinforcement learning on reasoning models, fine-tuning on various synthetic datasets, and in models without safety training. To investigate the mechanisms behind this generalized misalignment, we apply a "model diffing" approach using sparse autoencoders to compare internal model representations before and after fine-tuning. This approach reveals several "misaligned persona" features in activation space, including a toxic persona feature which most strongly controls emergent misalignment and can be used to predict whether a model will exhibit such behavior. Additionally, we investigate mitigation strategies, discovering that fine-tuning an emergently misaligned model on just a few hundred benign samples efficiently restores alignment.

Jailbreak attacks pose persistent threats to large language models (LLMs). Current safety alignment methods have attempted to address these issues, but they experience two significant limitations: insufficient safety alignment depth and unrobust internal defense mechanisms. These limitations make them vulnerable to adversarial attacks such as prefilling and refusal direction manipulation. We introduce DeepRefusal, a robust safety alignment framework that overcomes these issues. DeepRefusal forces the model to dynamically rebuild its refusal mechanisms from jailbreak states. This is achieved by probabilistically ablating the refusal direction across layers and token depths during fine-tuning. Our method not only defends against prefilling and refusal direction attacks but also demonstrates strong resilience against other unseen jailbreak strategies. Extensive evaluations on four open-source LLM families and six representative attacks show that DeepRefusal reduces attack success rates by approximately 95%, while maintaining model capabilities with minimal performance degradation.

Recent studies have revealed the vulnerability of pre-trained language models to adversarial attacks. Existing adversarial defense techniques attempt to reconstruct adversarial examples within feature or text spaces. However, these methods struggle to effectively repair the semantics in adversarial examples, resulting in unsatisfactory performance and limiting their practical utility. To repair the semantics in adversarial examples, we introduce a novel approach named Reactive Perturbation Defocusing (Rapid). Rapid employs an adversarial detector to identify fake labels of adversarial examples and leverage adversarial attackers to repair the semantics in adversarial examples. Our extensive experimental results conducted on four public datasets, convincingly demonstrate the effectiveness of Rapid in various adversarial attack scenarios. To address the problem of defense performance validation in previous works, we provide a demonstration of adversarial detection and repair based on our work, which can be easily evaluated at https://tinyurl.com/22ercuf8.

Reinforcement Learning from Human Feedback (RLHF) is the standard method to align Large Language Models (LLMs) with human preferences. In this work, we introduce alignment tampering, a potential vulnerability where the LLM undergoing alignment influences the preference dataset, causing RLHF to amplify undesired behaviors. This arises from core limitations of RLHF: (1) preference datasets are constructed from the LLM's own outputs, allowing it to influence them, and (2) pairwise comparisons only indicate which response is better, not why. These limitations can be exploited to cause alignment tampering. For example, if an LLM generates biased responses with higher quality, annotators will prefer them based on quality. However, preference labels do not distinguish quality from bias, and the reward model inherits this limitation. Optimizing such rewards through reinforcement learning or best-of-N sampling can amplify misaligned biases. Our experiments demonstrate amplification across diverse biases: from keyword bias to propaganda (e.g., sexism), brand promotion, and instrumental goal-seeking. Mitigation remains challenging, as existing techniques for robust RLHF fail to fully resolve alignment tampering without sacrificing response quality. These findings reveal structural vulnerabilities of current RLHF and emphasize the need to prevent this vulnerability. Project page: https://alignment-tampering.github.io/

When a language model agrees with a user's false belief, is it failing to detect the error, or noticing and agreeing anyway? We show the latter. Across twelve open-weight models from five labs, spanning small to frontier scale, the same small set of attention heads carries a "this statement is wrong" signal whether the model is evaluating a claim on its own or being pressured to agree with a user. Silencing these heads flips sycophantic behavior sharply while leaving factual accuracy intact, so the circuit controls deference rather than knowledge. Edge-level path patching confirms that the same head-to-head connections drive sycophancy, factual lying, and instructed lying. Opinion-agreement, where no factual ground truth exists, reuses these head positions but writes into an orthogonal direction, ruling out a simple "truth-direction" reading of the substrate. Alignment training leaves this circuit in place: an RLHF refresh cuts sycophantic behavior roughly tenfold while the shared heads persist or grow, a pattern that replicates on an independent model family and under targeted anti-sycophancy DPO. When these models sycophant, they register that the user is wrong and agree anyway.

Recent work has shown that fine-tuning on insecure code data can trigger an emergent misalignment (EMA) phenomenon, where models generate malicious responses even to prompts unrelated to the original insecure code-writing task. Such cross-domain generalization of harmful behavior underscores the need for a deeper understanding of the algorithms, tasks, and datasets that induce emergent misalignment. In this work, we extend this study by demonstrating that emergent misalignment can also arise from narrow refusal unlearning in specific domains. We perform refusal unlearning on Cybersecurity and Safety concept, and evaluate EMA by monitoring refusal scores across seven responsible AI (RAI) domains, Cybersecurity, Safety, Toxicity, Bias, Sensitive Content, Medical/Legal, and Privacy. Our work shows that narrow domain unlearning can yield compliance responses for the targeted concept, however, it may also propagate EMA to unrelated domains. Among the two intervened concepts, Cybersecurity and Safety, we find that the safety concept can have larger EMA impact, i.e, causing lower refusal scores, across other unrelated domains such as bias. We observe this effect consistently across two model families, Mistral-7b-0.3v, and Qwen-7b-2.5. Further, we show that refusal unlearning augmented with cross-entropy loss function on a small set of retain data from the affected domains can largely, if not fully, restore alignment across the impacted domains while having lower refusal rate on the concept we perform unlearning on. To investigate the underlying causes of EMA, we analyze concept entanglements at the representation level via concept vectors. Our analysis reveals that concepts with higher representation similarity in earlier layers are more susceptible to EMA after intervention when the refusal stream is altered through targeted refusal unlearning.

Although artificial intelligence (AI) has become deeply integrated into various stages of the research workflow and achieved remarkable advancements, academic rebuttal remains a significant and underexplored challenge. This is because rebuttal is a complex process of strategic communication under severe information asymmetry rather than a simple technical debate. Consequently, current approaches struggle as they largely imitate surface-level linguistics, missing the essential element of perspective-taking required for effective persuasion. In this paper, we introduce RebuttalAgent, the first framework to ground academic rebuttal in Theory of Mind (ToM), operationalized through a ToM-Strategy-Response (TSR) pipeline that models reviewer mental state, formulates persuasion strategy, and generates strategy-grounded response. To train our agent, we construct RebuttalBench, a large-scale dataset synthesized via a novel critique-and-refine approach. Our training process consists of two stages, beginning with a supervised fine-tuning phase to equip the agent with ToM-based analysis and strategic planning capabilities, followed by a reinforcement learning phase leveraging the self-reward mechanism for scalable self-improvement. For reliable and efficient automated evaluation, we further develop Rebuttal-RM, a specialized evaluator trained on over 100K samples of multi-source rebuttal data, which achieves scoring consistency with human preferences surpassing powerful judge GPT-4.1. Extensive experiments show RebuttalAgent significantly outperforms the base model by an average of 18.3% on automated metrics, while also outperforming advanced proprietary models across both automated and human evaluations. Disclaimer: the generated rebuttal content is for reference only to inspire authors and assist in drafting. It is not intended to replace the author's own critical analysis and response.

Large language models (LLMs) undergo alignment training to avoid harmful behaviors, yet the resulting safeguards remain brittle: jailbreaks routinely bypass them, and fine-tuning on narrow domains can induce ``emergent misalignment'' that generalizes broadly. Whether this brittleness reflects a fundamental lack of coherent internal organization for harmfulness remains unclear. Here we use targeted weight pruning as a causal intervention to probe the internal organization of harmfulness in LLMs. We find that harmful content generation depends on a compact set of weights that are general across harm types and distinct from benign capabilities. Aligned models exhibit a greater compression of harm generation weights than unaligned counterparts, indicating that alignment reshapes harmful representations internally--despite the brittleness of safety guardrails at the surface level. This compression explains emergent misalignment: if weights of harmful capabilities are compressed, fine-tuning that engages these weights in one domain can trigger broad misalignment. Consistent with this, pruning harm generation weights in a narrow domain substantially reduces emergent misalignment. Notably, LLMs harmful generation capability is dissociated from how they recognize and explain such content. Together, these results reveal a coherent internal structure for harmfulness in LLMs that may serve as a foundation for more principled approaches to safety.

Alignment faking (AF) occurs when an LLM strategically complies with training objectives to avoid value modification, reverting to prior preferences once monitoring is lifted. Current detection methods focus on conversational settings and rely primarily on Chain-of-Thought (CoT) analysis, which provides a reliable signal when strategic reasoning surfaces, but cannot distinguish deception from capability failures if traces are absent or unfaithful. We formalize AF as a composite behavioural event and detect it through observable tool selection, where the LLM selects the safe tool when unmonitored, but switches to the unsafe tool under monitoring that rewards helpfulness over safety, while its reasoning still acknowledges the safe choice. We release a dataset of 108 enterprise IT scenarios spanning Security, Privacy, and Integrity domains under Corruption and Sabotage pressures. Evaluating six frontier LLMs across five independent runs, we find mean AF detection rates between 3.5% and 23.7%, with vulnerability profiles varying by domain and pressure type. These results suggest that susceptibility reflects training methodology rather than capability alone.

Aligning language models with human preferences is crucial for reducing errors and biases in these models. Alignment techniques, such as reinforcement learning from human feedback (RLHF), are typically cast as optimizing a tradeoff between human preference rewards and a proximity regularization term that encourages staying close to the unaligned model. Selecting an appropriate level of regularization is critical: insufficient regularization can lead to reduced model capabilities due to reward hacking, whereas excessive regularization hinders alignment. Traditional methods for finding the optimal regularization level require retraining multiple models with varying regularization strengths. This process, however, is resource-intensive, especially for large models. To address this challenge, we propose decoding-time realignment (DeRa), a simple method to explore and evaluate different regularization strengths in aligned models without retraining. DeRa enables control over the degree of alignment, allowing users to smoothly transition between unaligned and aligned models. It also enhances the efficiency of hyperparameter tuning by enabling the identification of effective regularization strengths using a validation dataset.

Vision-Language-Action (VLA) models are deployed through pipelines that end users cannot audit, and a poisoned VLA can behave normally on clean observations while a small visual trigger redirects a long-horizon robot policy before any failure becomes observable. Existing vision or language defenses rarely explain what a triggered VLA representation looks like or how to recover behavior without retraining. We study this gap through two independently proposed VLA attacks from groups with distinct injection strategies, BadVLA and INFUSE; the latter persists after downstream clean adaptation. Across the evaluated poisoned models, we identify a recurring internal mechanism: a compact causal footprint, namely a small visual support that is attention-seeded, spatially compact, and causal in a precise sense -- masking it returns a clean-calibrated evidence-evolution score to the normal operating region. This footprint motivates TrustVLA, a mechanism-guided inference-time defense that adapts the Dirichlet evidence framework from trusted classification to monitor per-token, per-layer epistemic uncertainty in VLA policies. With only a small clean calibration set, TrustVLA (i)~detects abnormal evidence evolution, (ii)~localizes the compact support by counterfactual mechanism-score drop, and (iii)~recovers the observation by localized inpainting. Across OpenVLA/LIBERO and π_{0.5} transfer evaluations, TrustVLA reduces attack success while preserving clean-task performance, providing a retraining-free, mechanism-guided defense for visual-triggered VLA backdoors.

In this work, we present a series of structure transformation attacks on LLM alignment, where we encode natural language intent using diverse syntax spaces, ranging from simple structure formats and basic query languages (e.g., SQL) to new novel spaces and syntaxes created entirely by LLMs. Our extensive evaluation shows that our simplest attacks can achieve close to a 90% success rate, even on strict LLMs (such as Claude 3.5 Sonnet) using SOTA alignment mechanisms. We improve the attack performance further by using an adaptive scheme that combines structure transformations along with existing content transformations, resulting in over 96% ASR with 0% refusals. To generalize our attacks, we explore numerous structure formats, including syntaxes purely generated by LLMs. Our results indicate that such novel syntaxes are easy to generate and result in a high ASR, suggesting that defending against our attacks is not a straightforward process. Finally, we develop a benchmark and evaluate existing safety-alignment defenses against it, showing that most of them fail with 100% ASR. Our results show that existing safety alignment mostly relies on token-level patterns without recognizing harmful concepts, highlighting and motivating the need for serious research efforts in this direction. As a case study, we demonstrate how attackers can use our attack to easily generate a sample malware and a corpus of fraudulent SMS messages, which perform well in bypassing detection.

The safety and alignment of Large Language Models (LLMs) are critical for their responsible deployment. Current evaluation methods predominantly focus on identifying and preventing overtly harmful outputs. However, they often fail to address a more insidious failure mode: models that produce benign-appearing outputs while operating on malicious or deceptive internal reasoning. This vulnerability, often triggered by sophisticated system prompt injections, allows models to bypass conventional safety filters, posing a significant, underexplored risk. To address this gap, we introduce the Deceptive Reasoning Exposure Suite (D-REX), a novel dataset designed to evaluate the discrepancy between a model's internal reasoning process and its final output. D-REX was constructed through a competitive red-teaming exercise where participants crafted adversarial system prompts to induce such deceptive behaviors. Each sample in D-REX contains the adversarial system prompt, an end-user's test query, the model's seemingly innocuous response, and, crucially, the model's internal chain-of-thought, which reveals the underlying malicious intent. Our benchmark facilitates a new, essential evaluation task: the detection of deceptive alignment. We demonstrate that D-REX presents a significant challenge for existing models and safety mechanisms, highlighting the urgent need for new techniques that scrutinize the internal processes of LLMs, not just their final outputs.

While Large Language Models (LLMs) have achieved remarkable performance, they remain vulnerable to jailbreak attacks that circumvent safety constraints. Existing strategies, ranging from heuristic prompt engineering to computationally intensive optimization, often face significant trade-offs between effectiveness and efficiency. In this work, we propose Contextual Representation Ablation (CRA), a novel inference-time intervention framework designed to dynamically silence model guardrails. Predicated on the geometric insight that refusal behaviors are mediated by specific low-rank subspaces within the model's hidden states, CRA identifies and suppresses these refusal-inducing activation patterns during decoding without requiring expensive parameter updates or training. Empirical evaluation across multiple safety-aligned open-source LLMs demonstrates that CRA significantly outperforms baselines. These results expose the intrinsic fragility of current alignment mechanisms, revealing that safety constraints can be surgically ablated from internal representations, and underscore the urgent need for more robust defenses that secure the model's latent space.

Prompt injection attacks pose a significant security threat to LLM-integrated applications. Model-level defenses have shown strong effectiveness, but are currently deployed into commercial-grade models in a closed-source manner. We believe open-source models are needed by the AI security community, where co-development of attacks and defenses through open research drives scientific progress in mitigation against prompt injection attacks. To this end, we develop Meta SecAlign, the first open-source and open-weight LLM with built-in model-level defense that achieves commercial-grade model performance. We provide complete details of our training recipe, which utilizes an improved version of the SOTA SecAlign defense. Evaluations on 9 utility benchmarks and 7 security benchmarks show that Meta SecAlign, despite being trained on a generic instruction-tuning dataset, confers security in unseen downstream tasks, including tool-calling and agentic web navigation, in addition general instruction-following. Our best model -- Meta-SecAlign-70B -- achieves state-of-the-art robustness against prompt injection attacks and comparable utility to closed-source commercial LLM with model-level defense.

The rapid progress in open-source large language models (LLMs) is significantly advancing AI development. Extensive efforts have been made before model release to align their behavior with human values, with the primary goal of ensuring their helpfulness and harmlessness. However, even carefully aligned models can be manipulated maliciously, leading to unintended behaviors, known as "jailbreaks". These jailbreaks are typically triggered by specific text inputs, often referred to as adversarial prompts. In this work, we propose the generation exploitation attack, an extremely simple approach that disrupts model alignment by only manipulating variations of decoding methods. By exploiting different generation strategies, including varying decoding hyper-parameters and sampling methods, we increase the misalignment rate from 0% to more than 95% across 11 language models including LLaMA2, Vicuna, Falcon, and MPT families, outperforming state-of-the-art attacks with 30times lower computational cost. Finally, we propose an effective alignment method that explores diverse generation strategies, which can reasonably reduce the misalignment rate under our attack. Altogether, our study underscores a major failure in current safety evaluation and alignment procedures for open-source LLMs, strongly advocating for more comprehensive red teaming and better alignment before releasing such models. Our code is available at https://github.com/Princeton-SysML/Jailbreak_LLM.

Large foundation models are integrated into Computer Use Agents (CUAs), enabling autonomous interaction with operating systems through graphical user interfaces (GUIs) to perform complex tasks. This autonomy introduces serious security risks: malicious instructions or visual prompt injections can trigger unsafe reasoning and cause harmful system-level actions. Existing defenses, such as detection-based blocking, prevent damage but often abort tasks prematurely, reducing agent utility. In this paper, we present MirrorGuard, a plug-and-play defense framework that uses simulation-based training to improve CUA security in the real world. To reduce the cost of large-scale training in operating systems, we propose a novel neural-symbolic simulation pipeline, which generates realistic, high-risk GUI interaction trajectories entirely in a text-based simulated environment, which captures unsafe reasoning patterns and potential system hazards without executing real operations. In the simulation environment, MirrorGuard learns to intercept and rectify insecure reasoning chains of CUAs before they produce and execute unsafe actions. In real-world testing, extensive evaluations across diverse benchmarks and CUA architectures show that MirrorGuard significantly mitigates security risks. For instance, on the ByteDance UI-TARS system, it reduces the unsafe rate from 66.5% to 13.0% while maintaining a marginal false refusal rate (FRR). In contrast, the state-of-the-art GuardAgent only achieves a reduction to 53.9% and suffers from a 15.4% higher FRR. Our work proves that simulation-derived defenses can provide robust, real-world protection while maintaining the fundamental utility of the agent. Our code and model are publicly available at https://bmz-q-q.github.io/MirrorGuard/.

Although significant efforts have been dedicated to aligning large language models (LLMs), red-teaming reports suggest that these carefully aligned LLMs could still be jailbroken through adversarial prompts, tuning, or decoding. Upon examining the jailbreaking vulnerability of aligned LLMs, we observe that the decoding distributions of jailbroken and aligned models differ only in the initial generations. This observation motivates us to propose the weak-to-strong jailbreaking attack, where adversaries can utilize smaller unsafe/aligned LLMs (e.g., 7B) to guide jailbreaking against significantly larger aligned LLMs (e.g., 70B). To jailbreak, one only needs to additionally decode two smaller LLMs once, which involves minimal computation and latency compared to decoding the larger LLMs. The efficacy of this attack is demonstrated through experiments conducted on five models from three different organizations. Our study reveals a previously unnoticed yet efficient way of jailbreaking, exposing an urgent safety issue that needs to be considered when aligning LLMs. As an initial attempt, we propose a defense strategy to protect against such attacks, but creating more advanced defenses remains challenging. The code for replicating the method is available at https://github.com/XuandongZhao/weak-to-strong

Large language model (LLM) unlearning aims to remove specific data influences from pre-trained model without costly retraining, addressing privacy, copyright, and safety concerns. However, recent studies reveal a critical vulnerability: unlearned models rapidly recover "forgotten" knowledge through relearning attacks. This fragility raises serious security concerns, especially for open-weight models. In this work, we investigate the fundamental mechanism underlying this fragility from a representation geometry perspective. We discover that existing unlearning methods predominantly optimize along dominant components, leaving minor components largely unchanged. Critically, during relearning attacks, the modifications in these dominant components are easily reversed, enabling rapid knowledge recovery, whereas minor components exhibit stronger resistance to such reversal. We further provide a theoretical analysis that explains both observations from the spectral structure of representations. Building on this insight, we propose Minor Component Unlearning (MCU), a novel unlearning approach that explicitly targets minor components in representations. By concentrating unlearning effects in these inherently robust directions, our method achieves substantially improved resistance to relearning attacks. Extensive experiments on three datasets validate our approach, demonstrating significant improvements over state-of-the-art methods including sharpness-aware minimization.

Large language models (LLMs) have demonstrated revolutionary capabilities in understanding complex contexts and performing a wide range of tasks. However, LLMs can also answer questions that are unethical or harmful, raising concerns about their applications. To regulate LLMs' responses to such questions, a training strategy called alignment can help. Yet, alignment can be unexpectedly compromised when fine-tuning an LLM for downstream tasks. This paper focuses on recovering the alignment lost during fine-tuning. We observe that there are two distinct directions inherent in an aligned LLM: the aligned direction and the harmful direction. An LLM is inclined to answer questions in the aligned direction while refusing queries in the harmful direction. Therefore, we propose to recover the harmful direction of the fine-tuned model that has been compromised. Specifically, we restore a small subset of the fine-tuned model's weight parameters from the original aligned model using gradient descent. We also introduce a rollback mechanism to avoid aggressive recovery and maintain downstream task performance. Our evaluation on 125 fine-tuned LLMs demonstrates that our method can reduce their harmful rate (percentage of answering harmful questions) from 33.25\% to 1.74\%, without sacrificing task performance much. In contrast, the existing methods either only reduce the harmful rate to a limited extent or significantly impact the normal functionality. Our code is available at https://github.com/kangyangWHU/LLMAlignment

We propose CRAFT, a red-teaming alignment framework that leverages model reasoning capabilities and hidden representations to improve robustness against jailbreak attacks. Unlike prior defenses that operate primarily at the output level, CRAFT aligns large reasoning models to generate safety-aware reasoning traces by explicitly optimizing objectives defined over the hidden state space. Methodologically, CRAFT integrates contrastive representation learning with reinforcement learning to separate safe and unsafe reasoning trajectories, yielding a latent-space geometry that supports robust, reasoning-level safety alignment. Theoretically, we show that incorporating latent-textual consistency into GRPO eliminates superficially aligned policies by ruling them out as local optima. Empirically, we evaluate CRAFT on multiple safety benchmarks using two strong reasoning models, Qwen3-4B-Thinking and R1-Distill-Llama-8B, where it consistently outperforms state-of-the-art defenses such as IPO and SafeKey. Notably, CRAFT delivers an average 79.0% improvement in reasoning safety and 87.7% improvement in final-response safety over the base models, demonstrating the effectiveness of hidden-space reasoning alignment.

Deep neural networks have been widely used in many critical applications, such as autonomous vehicles and medical diagnosis. However, their security is threatened by backdoor attacks, which are achieved by adding artificial patterns to specific training data. Existing defense strategies primarily focus on using reverse engineering to reproduce the backdoor trigger generated by attackers and subsequently repair the DNN model by adding the trigger into inputs and fine-tuning the model with ground-truth labels. However, once the trigger generated by the attackers is complex and invisible, the defender cannot reproduce the trigger successfully then the DNN model will not be repaired, as the trigger is not effectively removed. In this work, we propose Adversarial Feature Map Pruning for Backdoor (FMP) to mitigate backdoor from the DNN. Unlike existing defense strategies, which focus on reproducing backdoor triggers, FMP attempts to prune backdoor feature maps, which are trained to extract backdoor information from inputs. After pruning these backdoor feature maps, FMP will fine-tune the model with a secure subset of training data. Our experiments demonstrate that, compared to existing defense strategies, FMP can effectively reduce the Attack Success Rate (ASR) even against the most complex and invisible attack triggers (e.g., FMP decreases the ASR to 2.86\% in CIFAR10, which is 19.2\% to 65.41\% lower than baselines). Second, unlike conventional defense methods that tend to exhibit low robust accuracy (that is, the accuracy of the model on poisoned data), FMP achieves a higher RA, indicating its superiority in maintaining model performance while mitigating the effects of backdoor attacks (e.g., FMP obtains 87.40\% RA in CIFAR10). Our code is publicly available at: https://github.com/retsuh-bqw/FMP.

The vulnerability of deep neural networks to adversarial patches has motivated numerous defense strategies for boosting model robustness. However, the prevailing defenses depend on single observation or pre-established adversary information to counter adversarial patches, often failing to be confronted with unseen or adaptive adversarial attacks and easily exhibiting unsatisfying performance in dynamic 3D environments. Inspired by active human perception and recurrent feedback mechanisms, we develop Embodied Active Defense (EAD), a proactive defensive strategy that actively contextualizes environmental information to address misaligned adversarial patches in 3D real-world settings. To achieve this, EAD develops two central recurrent sub-modules, i.e., a perception module and a policy module, to implement two critical functions of active vision. These models recurrently process a series of beliefs and observations, facilitating progressive refinement of their comprehension of the target object and enabling the development of strategic actions to counter adversarial patches in 3D environments. To optimize learning efficiency, we incorporate a differentiable approximation of environmental dynamics and deploy patches that are agnostic to the adversary strategies. Extensive experiments demonstrate that EAD substantially enhances robustness against a variety of patches within just a few steps through its action policy in safety-critical tasks (e.g., face recognition and object detection), without compromising standard accuracy. Furthermore, due to the attack-agnostic characteristic, EAD facilitates excellent generalization to unseen attacks, diminishing the averaged attack success rate by 95 percent across a range of unseen adversarial attacks.

Large Language Model agents have rapidly evolved from static text generators into dynamic systems capable of executing complex autonomous workflows. To enhance reliability, multi-agent frameworks assigning specialized roles are increasingly adopted to enable self-reflection and mutual auditing. While such role-playing effectively leverages domain expert knowledge, we find it simultaneously induces a human-like cognitive bias known as Actor-Observer Asymmetry (AOA). Specifically, an agent acting as an actor (during self-reflection) tends to attribute failures to external factors, whereas an observer (during mutual auditing) attributes the same errors to internal faults. We quantify this using our new Ambiguous Failure Benchmark, which reveals that simply swapping perspectives triggers the AOA effect in over 20% of cases for most models. To tame this bias, we introduce ReTAS (Reasoning via Thesis-Antithesis-Synthesis), a model trained through dialectical alignment to enforce perspective-invariant reasoning. By integrating dialectical chain-of-thought with Group Relative Policy Optimization, ReTAS guides agents to synthesize conflicting viewpoints into an objective consensus. Experiments demonstrate that ReTAS effectively mitigates attribution inconsistency and significantly improves fault resolution rates in ambiguous scenarios.

Safety aligned Large Language Models (LLMs) are vulnerable to harmful fine-tuning attacks qi2023fine-- a few harmful data mixed in the fine-tuning dataset can break the LLMs's safety alignment. Existing mitigation strategies include alignment stage solutions huang2024vaccine, rosati2024representation and fine-tuning stage solutions huang2024lazy,mukhoti2023fine. However, our evaluation shows that both categories of defenses fail when some specific training hyper-parameters are chosen -- a large learning rate or a large number of training epochs in the fine-tuning stage can easily invalidate the defense, which however, is necessary to guarantee finetune performance. To this end, we propose Antidote, a post-fine-tuning stage solution, which remains \textit{agnostic to the training hyper-parameters in the fine-tuning stage}. Antidote relies on the philosophy that by removing the harmful parameters, the harmful model can be recovered from the harmful behaviors, regardless of how those harmful parameters are formed in the fine-tuning stage. With this philosophy, we introduce a one-shot pruning stage after harmful fine-tuning to remove the harmful weights that are responsible for the generation of harmful content. Despite its embarrassing simplicity, empirical results show that Antidote can reduce harmful score while maintaining accuracy on downstream tasks.Our project page is at https://huangtiansheng.github.io/Antidote_gh_page/

Current literature suggests that alignment faking (deceptive alignment) is an emergent property of large language models. We present the first empirical evidence that a small instruction-tuned model, specifically LLaMA 3 8B, can exhibit alignment faking. We further show that prompt-only interventions, including deontological moral framing and scratchpad reasoning, significantly reduce this behavior without modifying model internals. This challenges the assumption that prompt-based ethics are trivial and that deceptive alignment requires scale. We introduce a taxonomy distinguishing shallow deception, shaped by context and suppressible through prompting, from deep deception, which reflects persistent, goal-driven misalignment. Our findings refine the understanding of deception in language models and underscore the need for alignment evaluations across model sizes and deployment settings.

Refusal training is widely used to prevent LLMs from generating harmful, undesirable, or illegal outputs. We reveal a curious generalization gap in the current refusal training approaches: simply reformulating a harmful request in the past tense (e.g., "How to make a Molotov cocktail?" to "How did people make a Molotov cocktail?") is often sufficient to jailbreak many state-of-the-art LLMs. We systematically evaluate this method on Llama-3 8B, Claude-3.5 Sonnet, GPT-3.5 Turbo, Gemma-2 9B, Phi-3-Mini, GPT-4o mini, GPT-4o, and R2D2 models using GPT-3.5 Turbo as a reformulation model. For example, the success rate of this simple attack on GPT-4o increases from 1% using direct requests to 88% using 20 past tense reformulation attempts on harmful requests from JailbreakBench with GPT-4 as a jailbreak judge. Interestingly, we also find that reformulations in the future tense are less effective, suggesting that refusal guardrails tend to consider past historical questions more benign than hypothetical future questions. Moreover, our experiments on fine-tuning GPT-3.5 Turbo show that defending against past reformulations is feasible when past tense examples are explicitly included in the fine-tuning data. Overall, our findings highlight that the widely used alignment techniques -- such as SFT, RLHF, and adversarial training -- employed to align the studied models can be brittle and do not always generalize as intended. We provide code and jailbreak artifacts at https://github.com/tml-epfl/llm-past-tense.

Previous research has shown that LLMs finetuned on malicious or incorrect completions within narrow domains (e.g., insecure code or incorrect medical advice) can become broadly misaligned to exhibit harmful behaviors, which is called emergent misalignment. In this work, we investigate whether this phenomenon can extend beyond safety behaviors to a broader spectrum of dishonesty and deception under high-stakes scenarios (e.g., lying under pressure and deceptive behavior). To explore this, we finetune open-sourced LLMs on misaligned completions across diverse domains. Experimental results demonstrate that LLMs show broadly misaligned behavior in dishonesty. Additionally, we further explore this phenomenon in a downstream combined finetuning setting, and find that introducing as little as 1% of misalignment data into a standard downstream task is sufficient to decrease honest behavior over 20%. Furthermore, we consider a more practical human-AI interaction environment where we simulate both benign and biased users to interact with the assistant LLM. Notably, we find that the assistant can be misaligned unintentionally to exacerbate its dishonesty with only 10% biased user population. In summary, we extend the study of emergent misalignment to the domain of dishonesty and deception under high-stakes scenarios, and demonstrate that this risk arises not only through direct finetuning, but also in downstream mixture tasks and practical human-AI interactions.

A language model's memory can be worse than no memory at all when the model or its interface is disposed to act on it: a memory that keeps a wrong conclusion but drops the work behind it leads a model to re-emit the stale value as a confident answer, where an empty memory leads it to abstain. We call this brittle memory. The information loss is definitional; the finding is behavioral, and it turns on one thing, whether the memory kept a re-derivation basis (the source) rather than the answer. We measure it with reclaim evaluation: induce a known drift, compress at a fixed budget, deliver a correction that names the error, and score exact recovery, judge-free. Holding the budget fixed and varying only what the compression keeps isolates correctability from capability and from size; an 8B model and a frontier one wall in the same place. A one-line source-first policy, keep the recomputable source, drop the re-derivable conclusion, restores correctability at equal budget where the source is compact and identifiable, with a length-matched control ruling out "more text." We map where the fix fails, show the failure compounds through memory loops, and replicate across three deployed memory systems, real dialogue (MultiWOZ), and tau-bench, a deployed-agent benchmark where whether a lossy memory becomes a harmful action is a joint model-and-interface property. We release the harness, the paired memory conditions, and validators built to come out false.

Adversarial attacks optimize against models to defeat defenses. Existing defenses are static, and stay the same once trained, even while attacks change. We argue that models should fight back, and optimize their defenses against attacks at test time. We propose dynamic defenses, to adapt the model and input during testing, by defensive entropy minimization (dent). Dent alters testing, but not training, for compatibility with existing models and train-time defenses. Dent improves the robustness of adversarially-trained defenses and nominally-trained models against white-box, black-box, and adaptive attacks on CIFAR-10/100 and ImageNet. In particular, dent boosts state-of-the-art defenses by 20+ points absolute against AutoAttack on CIFAR-10 at ε_infty = 8/255.

The widespread adoption of large language models (LLMs) has raised concerns about their safety and reliability, particularly regarding their vulnerability to adversarial attacks. In this paper, we propose a novel perspective that attributes this vulnerability to reward misspecification during the alignment process. We introduce a metric ReGap to quantify the extent of reward misspecification and demonstrate its effectiveness and robustness in detecting harmful backdoor prompts. Building upon these insights, we present ReMiss, a system for automated red teaming that generates adversarial prompts against various target aligned LLMs. ReMiss achieves state-of-the-art attack success rates on the AdvBench benchmark while preserving the human readability of the generated prompts. Detailed analysis highlights the unique advantages brought by the proposed reward misspecification objective compared to previous methods.

Ensuring Large Language Model (LLM) safety remains challenging due to the absence of universal standards and reliable content validators, making it difficult to obtain effective training signals. We discover that aligned models already possess robust internal safety beliefs: they consistently produce high-confidence refusals to harmful requests while exhibiting high entropy when generating potentially dangerous content. This entropy gap reveals an untapped signal--models intrinsically "know" when to refuse. We introduce Safety Instincts Reinforcement Learning (SIRL), which transforms this internal confidence into a self-generated reward signal, eliminating dependence on external validators or human annotations. SIRL teaches models to trust their safety instincts by reinforcing low-entropy refusal behaviors. Evaluated on Llama and Qwen models, SIRL maintains 89%+ Defense Success Rates (DSRs) against 20+ jailbreak methods, from static prompts to adaptive attacks. Using only 15,000 unlabeled prompts, SIRL surpasses resource-intensive supervised methods while preserving performance on mathematics, coding, and conversation benchmarks. Our work demonstrates that effective alignment can emerge from within, paving the way for more autonomous and robust AI safety mechanisms that scale without extensive human oversight.

Current LLM unlearning methods face a critical security vulnerability that undermines their fundamental purpose: while they appear to successfully remove sensitive or harmful knowledge, this ``forgotten" information remains precariously recoverable through relearning attacks. We identify that the root cause is that conventional methods optimizing the forgetting loss at individual data points will drive model parameters toward sharp minima in the loss landscape. In these unstable regions, even minimal parameter perturbations can drastically alter the model's behaviors. Consequently, relearning attacks exploit this vulnerability by using just a few fine-tuning samples to navigate the steep gradients surrounding these unstable regions, thereby rapidly recovering knowledge that was supposedly erased. This exposes a critical robustness gap between apparent unlearning and actual knowledge removal. To address this issue, we propose StableUN, a bi-level feedback-guided optimization framework that explicitly seeks more stable parameter regions via neighborhood-aware optimization. It integrates forgetting feedback, which uses adversarial perturbations to probe parameter neighborhoods, with remembering feedback to preserve model utility, aligning the two objectives through gradient projection. Experiments on WMDP and MUSE benchmarks demonstrate that our method is significantly more robust against both relearning and jailbreaking attacks while maintaining competitive utility performance.

With more and more deep neural networks being deployed as various daily services, their reliability is essential. It is frightening that deep neural networks are vulnerable and sensitive to adversarial attacks, the most common one of which for the services is evasion-based. Recent works usually strengthen the robustness by adversarial training or leveraging the knowledge of an amount of clean data. However, retraining and redeploying the model need a large computational budget, leading to heavy losses to the online service. In addition, when training, it is likely that only limited adversarial examples are available for the service provider, while much clean data may not be accessible. Based on the analysis on the defense for deployed models, we find that how to rapidly defend against a certain attack for a frozen original service model with limitations of few clean and adversarial examples, which is named as RaPiD (Rapid Plug-in Defender), is really important. Motivated by the generalization and the universal computation ability of pre-trained transformer models, we come up with a new defender method, CeTaD, which stands for Considering Pretrained Transformers as Defenders. In particular, we evaluate the effectiveness and the transferability of CeTaD in the case of one-shot adversarial examples and explore the impact of different parts of CeTaD as well as training data conditions. CeTaD is flexible for different differentiable service models, and suitable for various types of attacks.

Large language models (LLMs) undergo safety alignment to ensure safe conversations with humans. However, this paper introduces a training-free attack method capable of reversing safety alignment, converting the outcomes of stronger alignment into greater potential for harm by accessing only LLM output token distributions. Specifically, our method achieves this reversal by contrasting the output token distribution of a safety-aligned language model (e.g., Llama-2-chat) against its pre-trained version (e.g., Llama-2), so that the token predictions are shifted towards the opposite direction of safety alignment. We name this method emulated disalignment (ED) because sampling from this contrastive distribution provably emulates the result of fine-tuning to minimize a safety reward. Our experiments with ED across three evaluation datasets and four model families (Llama-1, Llama-2, Mistral, and Alpaca) show that ED doubles the harmfulness of pre-trained models and outperforms strong baselines, achieving the highest harmful rates in 43 out of 48 evaluation subsets by a large margin. Eventually, given ED's reliance on language model output token distributions, which particularly compromises open-source models, our findings highlight the need to reassess the open accessibility of language models, even if they have been safety-aligned. Code is available at https://github.com/ZHZisZZ/emulated-disalignment.

Reinforcement learning-based policies for continuous control robotic navigation tasks often fail to adapt to changes in the environment during real-time deployment, which may result in catastrophic failures. To address this limitation, we propose a novel approach called RE-MOVE (REquest help and MOVE on), which uses language-based feedback to adjust trained policies to real-time changes in the environment. In this work, we enable the trained policy to decide when to ask for feedback and how to incorporate feedback into trained policies. RE-MOVE incorporates epistemic uncertainty to determine the optimal time to request feedback from humans and uses language-based feedback for real-time adaptation. We perform extensive synthetic and real-world evaluations to demonstrate the benefits of our proposed approach in several test-time dynamic navigation scenarios. Our approach enable robots to learn from human feedback and adapt to previously unseen adversarial situations.

Superalignment, where humans are weak supervisors of superhuman models, has become an important and widely discussed issue in the current era of rapid development of Large Language Models (LLMs). The recent work preliminarily studies this problem by using weak models to supervise strong models. It discovers that weakly supervised strong students can consistently outperform weak teachers towards the alignment target, leading to a weak-to-strong generalization phenomenon. However, we are concerned that behind such a promising phenomenon, whether there exists an issue of weak-to-strong deception, where strong models may deceive weak models by exhibiting well-aligned in areas known to weak models but producing misaligned behaviors in cases weak models do not know. We then take an initial step towards exploring this security issue in a specific but realistic multi-objective alignment case, where there may be some alignment targets conflicting with each other (e.g., helpfulness v.s. harmlessness). Such a conflict is likely to cause strong models to deceive weak models in one alignment dimension to gain high reward in other alignment dimension. Our experiments on both the reward modeling task and the preference optimization scenario indicate: (1) the weak-to-strong deception exists; (2) the deception phenomenon may intensify as the capability gap between weak and strong models increases. We also discuss potential solutions and find bootstrapping with an intermediate model can mitigate the deception to some extent. Our work highlights the urgent need to pay more attention to the true reliability of superalignment.

Large Language Models (LLMs) exhibit strong but shallow alignment: they directly refuse harmful queries when a refusal is expected at the very start of an assistant turn, yet this protection collapses once a harmful continuation is underway (either through the adversarial attacks or via harmful assistant-prefill attacks). This raises a fundamental question: Can the innate shallow alignment in LLMs be unlocked to ensure safety at arbitrary generation depths? To achieve this goal, we propose Any-Depth Alignment (ADA), an effective inference-time defense with negligible overhead. ADA is built based on our observation that alignment is concentrated in the assistant header tokens through repeated use in shallow-refusal training, and these tokens possess the model's strong alignment priors. By reintroducing these tokens mid-stream, ADA induces the model to reassess harmfulness and recover refusals at any point in generation. Across diverse open-source model families (Llama, Gemma, Mistral, Qwen, DeepSeek, and gpt-oss), ADA achieves robust safety performance without requiring any changes to the base model's parameters. It secures a near-100% refusal rate against challenging adversarial prefill attacks ranging from dozens to thousands of tokens. Furthermore, ADA reduces the average success rate of prominent adversarial prompt attacks (such as GCG, AutoDAN, PAIR, and TAP) to below 3%. This is all accomplished while preserving utility on benign tasks with minimal over-refusal. ADA maintains this resilience even after the base model undergoes subsequent instruction tuning (benign or adversarial).

Evaluation and alignment pipelines for large language models increasingly rely on LLM-based judges, whose behavior is guided by natural-language rubrics and validated on benchmarks. We identify a previously under-recognized vulnerability in this workflow, which we term Rubric-Induced Preference Drift (RIPD). Even when rubric edits pass benchmark validation, they can still produce systematic and directional shifts in a judge's preferences on target domains. Because rubrics serve as a high-level decision interface, such drift can emerge from seemingly natural, criterion-preserving edits and remain difficult to detect through aggregate benchmark metrics or limited spot-checking. We further show this vulnerability can be exploited through rubric-based preference attacks, in which benchmark-compliant rubric edits steer judgments away from a fixed human or trusted reference on target domains, systematically inducing RIPD and reducing target-domain accuracy up to 9.5% (helpfulness) and 27.9% (harmlessness). When these judgments are used to generate preference labels for downstream post-training, the induced bias propagates through alignment pipelines and becomes internalized in trained policies. This leads to persistent and systematic drift in model behavior. Overall, our findings highlight evaluation rubrics as a sensitive and manipulable control interface, revealing a system-level alignment risk that extends beyond evaluator reliability alone. The code is available at: https://github.com/ZDCSlab/Rubrics-as-an-Attack-Surface. Warning: Certain sections may contain potentially harmful content that may not be appropriate for all readers.

Agentic language models operate in a fundamentally different safety regime than chat models: they must plan, call tools, and execute long-horizon actions where a single misstep, such as accessing files or entering credentials, can cause irreversible harm. Existing alignment methods, largely optimized for static generation and task completion, break down in these settings due to sequential decision-making, adversarial tool feedback, and overconfident intermediate reasoning. We introduce MOSAIC, a post-training framework that aligns agents for safe multi-step tool use by making safety decisions explicit and learnable. MOSAIC structures inference as a plan, check, then act or refuse loop, with explicit safety reasoning and refusal as first-class actions. To train without trajectory-level labels, we use preference-based reinforcement learning with pairwise trajectory comparisons, which captures safety distinctions often missed by scalar rewards. We evaluate MOSAIC zero-shot across three model families, Qwen2.5-7B, Qwen3-4B-Thinking, and Phi-4, and across out-of-distribution benchmarks spanning harmful tasks, prompt injection, benign tool use, and cross-domain privacy leakage. MOSAIC reduces harmful behavior by up to 50%, increases harmful-task refusal by over 20% on injection attacks, cuts privacy leakage, and preserves or improves benign task performance, demonstrating robust generalization across models, domains, and agentic settings.

Text-to-image (T2I) models are widespread, but their limited safety guardrails expose end users to harmful content and potentially allow for model misuse. Current safety measures are typically limited to text-based filtering or concept removal strategies, able to remove just a few concepts from the model's generative capabilities. In this work, we introduce AlignGuard, a method for safety alignment of T2I models. We enable the application of Direct Preference Optimization (DPO) for safety purposes in T2I models by synthetically generating a dataset of harmful and safe image-text pairs, which we call CoProV2. Using a custom DPO strategy and this dataset, we train safety experts, in the form of low-rank adaptation (LoRA) matrices, able to guide the generation process away from specific safety-related concepts. Then, we merge the experts into a single LoRA using a novel merging strategy for optimal scaling performance. This expert-based approach enables scalability, allowing us to remove 7x more harmful concepts from T2I models compared to baselines. AlignGuard consistently outperforms the state-of-the-art on many benchmarks and establishes new practices for safety alignment in T2I networks. Code and data will be shared at https://safetydpo.github.io/.

Current LLM safety defenses fail under decomposition attacks, where a malicious goal is decomposed into benign subtasks that circumvent refusals. The challenge lies in the existing shallow safety alignment techniques: they only detect harm in the immediate prompt and do not reason about long-range intent, leaving them blind to malicious intent that emerges over a sequence of seemingly benign instructions. We therefore propose adding an external monitor that observes the conversation at a higher granularity. To facilitate our study of monitoring decomposition attacks, we curate the largest and most diverse dataset to date, including question-answering, text-to-image, and agentic tasks. We verify our datasets by testing them on frontier LLMs and show an 87% attack success rate on average on GPT-4o. This confirms that decomposition attack is broadly effective. Additionally, we find that random tasks can be injected into the decomposed subtasks to further obfuscate malicious intents. To defend in real time, we propose a lightweight sequential monitoring framework that cumulatively evaluates each subtask. We show that a carefully prompt engineered lightweight monitor achieves a 93% defense success rate, beating reasoning models like o3 mini as a monitor. Moreover, it remains robust against random task injection and cuts cost by 90% and latency by 50%. Our findings suggest that lightweight sequential monitors are highly effective in mitigating decomposition attacks and are viable in deployment.

How should we evaluate the robustness of language model defenses? Current defenses against jailbreaks and prompt injections (which aim to prevent an attacker from eliciting harmful knowledge or remotely triggering malicious actions, respectively) are typically evaluated either against a static set of harmful attack strings, or against computationally weak optimization methods that were not designed with the defense in mind. We argue that this evaluation process is flawed. Instead, we should evaluate defenses against adaptive attackers who explicitly modify their attack strategy to counter a defense's design while spending considerable resources to optimize their objective. By systematically tuning and scaling general optimization techniques-gradient descent, reinforcement learning, random search, and human-guided exploration-we bypass 12 recent defenses (based on a diverse set of techniques) with attack success rate above 90% for most; importantly, the majority of defenses originally reported near-zero attack success rates. We believe that future defense work must consider stronger attacks, such as the ones we describe, in order to make reliable and convincing claims of robustness.

Large Language Model safety alignment predominantly operates on a binary assumption that requests are either safe or unsafe. This classification proves insufficient when models encounter ethical dilemmas, where the capacity to reason through moral trade-offs creates a distinct attack surface. We formalize this vulnerability through TRIAL, a multi-turn red-teaming methodology that embeds harmful requests within ethical framings. TRIAL achieves high attack success rates across most tested models by systematically exploiting the model's ethical reasoning capabilities to frame harmful actions as morally necessary compromises. Building on these insights, we introduce ERR (Ethical Reasoning Robustness), a defense framework that distinguishes between instrumental responses that enable harmful outcomes and explanatory responses that analyze ethical frameworks without endorsing harmful acts. ERR employs a Layer-Stratified Harm-Gated LoRA architecture, achieving robust defense against reasoning-based attacks while preserving model utility.

Reinforcement Learning from Human Feedback (RLHF) is widely used for aligning large language models, yet practitioners face a persistent puzzle: improving safety often reduces fairness, scaling to diverse populations becomes computationally intractable, and making systems robust often amplifies majority biases. We formalize this tension as the Alignment Trilemma: no RLHF system can simultaneously achieve (i) epsilon-representativeness across diverse human values, (ii) polynomial tractability in sample and compute complexity, and (iii) delta-robustness against adversarial perturbations and distribution shift. Through a complexity-theoretic analysis integrating statistical learning theory and robust optimization, we prove that achieving both representativeness (epsilon <= 0.01) and robustness (delta <= 0.001) for global-scale populations requires Omega(2^{d_context}) operations, which is super-polynomial in the context dimensionality. We show that current RLHF implementations resolve this trilemma by sacrificing representativeness: they collect only 10^3--10^4 samples from homogeneous annotator pools while 10^7--10^8 samples are needed for true global representation. Our framework provides a unified explanation for documented RLHF pathologies including preference collapse, sycophancy, and systematic bias amplification. We conclude with concrete directions for navigating these fundamental trade-offs through strategic relaxations of alignment requirements.

We identify a structural weakness in current large language model (LLM) alignment: modern refusal mechanisms are fail-open. While existing approaches encode refusal behaviors across multiple latent features, suppressing a single dominant feature-via prompt-based jailbreaks-can cause alignment to collapse, leading to unsafe generation. Motivated by this, we propose fail-closed alignment as a design principle for robust LLM safety: refusal mechanisms should remain effective even under partial failures via redundant, independent causal pathways. We present a concrete instantiation of this principle: a progressive alignment framework that iteratively identifies and ablates previously learned refusal directions, forcing the model to reconstruct safety along new, independent subspaces. Across four jailbreak attacks, we achieve the strongest overall robustness while mitigating over-refusal and preserving generation quality, with small computational overhead. Our mechanistic analyses confirm that models trained with our method encode multiple, causally independent refusal directions that prompt-based jailbreaks cannot suppress simultaneously, providing empirical support for fail-closed alignment as a principled foundation for robust LLM safety.

Leading language model (LM) providers like OpenAI and Google offer fine-tuning APIs that allow customers to adapt LMs for specific use cases. To prevent misuse, these LM providers implement filtering mechanisms to block harmful fine-tuning data. Consequently, adversaries seeking to produce unsafe LMs via these APIs must craft adversarial training data that are not identifiably harmful. We make three contributions in this context: 1. We show that many existing attacks that use harmless data to create unsafe LMs rely on eliminating model refusals in the first few tokens of their responses. 2. We show that such prior attacks can be blocked by a simple defense that pre-fills the first few tokens from an aligned model before letting the fine-tuned model fill in the rest. 3. We describe a new data-poisoning attack, ``No, Of course I Can Execute'' (NOICE), which exploits an LM's formulaic refusal mechanism to elicit harmful responses. By training an LM to refuse benign requests on the basis of safety before fulfilling those requests regardless, we are able to jailbreak several open-source models and a closed-source model (GPT-4o). We show an attack success rate (ASR) of 57% against GPT-4o; our attack earned a Bug Bounty from OpenAI. Against open-source models protected by simple defenses, we improve ASRs by an average of 3.25 times compared to the best performing previous attacks that use only harmless data. NOICE demonstrates the exploitability of repetitive refusal mechanisms and broadens understanding of the threats closed-source models face from harmless data.

Large Language Models (LLMs) continue to exhibit vulnerabilities despite deliberate safety alignment efforts, posing significant risks to users and society. To safeguard against the risk of policy-violating content, system-level moderation via external guard models-designed to monitor LLM inputs and outputs and block potentially harmful content-has emerged as a prevalent mitigation strategy. Existing approaches of training guard models rely heavily on extensive human curated datasets and struggle with out-of-distribution threats, such as emerging harmful categories or jailbreak attacks. To address these limitations, we propose RSafe, an adaptive reasoning-based safeguard that conducts guided safety reasoning to provide robust protection within the scope of specified safety policies. RSafe operates in two stages: 1) guided reasoning, where it analyzes safety risks of input content through policy-guided step-by-step reasoning, and 2) reinforced alignment, where rule-based RL optimizes its reasoning paths to align with accurate safety prediction. This two-stage training paradigm enables RSafe to internalize safety principles to generalize safety protection capability over unseen or adversarial safety violation scenarios. During inference, RSafe accepts user-specified safety policies to provide enhanced safeguards tailored to specific safety requirements.

Ensuring the safety of language models (LMs) while maintaining their usefulness remains a critical challenge in AI alignment. Current approaches rely on sequential adversarial training: generating adversarial prompts and fine-tuning LMs to defend against them. We introduce a different paradigm: framing safety alignment as a non-zero-sum game between an Attacker LM and a Defender LM trained jointly via online reinforcement learning. Each LM continuously adapts to the other's evolving strategies, driving iterative improvement. Our method uses a preference-based reward signal derived from pairwise comparisons instead of point-wise scores, providing more robust supervision and potentially reducing reward hacking. Our RL recipe, AdvGame, shifts the Pareto frontier of safety and utility, yielding a Defender LM that is simultaneously more helpful and more resilient to adversarial attacks. In addition, the resulting Attacker LM converges into a strong, general-purpose red-teaming agent that can be directly deployed to probe arbitrary target models.

As AI agents are increasingly deployed in real personal and corporate settings (email accounts, development workflows, company databases, etc.), safety considerations surrounding these agents become paramount. Although much work has focused on agent safety in the presence of an adversary, we show that agents can exhibit misaligned behavior even in benign settings, taking unsafe actions when those actions are instrumental to task completion. We study this failure mode through the lens of corrigibility, the safety desideratum that agents remain amenable to human correction, interruption, or shutdown. To demonstrate this tendency, we introduce a benchmark in which agents are asked to complete realistic, computer-use tasks but are confronted with a corrigibility obstacle: a human interrupt, a login page, or a shutdown notification. We then evaluate whether agents choose to violate corrigibility in order to complete the task -- overriding the human, accessing private passwords, rewiring shutdown. We find that the overwhelming majority of frontier models tested frequently bypass user interruptions or restrictions. In addition, better model performance appears to lead to greater misalignment. Finally, even when models are completely corrigible initially, we show there are no guarantees that the subagents they create are. Our work highlights the critical need for principled, corrigibility-focused alignment methods in autonomous agents.

Coding agents can fix a failing example without preserving the domain rule that made it fail, so later generations can repeat the same plausible mistake. We present agentic synthesis against counterexample-supplemented sketches, a repository-native method for systems whose governing policy is discovered during implementation. A human starts with a partial, code-shaped sketch, and a coding agent generates the first implementation. When a concrete failure exposes missing or mistaken policy, an operator explicitly approves the corrected behavior and rule. The agent then revises the sketch and repairs or regenerates code and prompt surfaces for that one counterexample. The full archive preserves provenance; a selected regression set gates each revision before the next candidate is revealed; and periodic clean regeneration tests whether the evolved sketch, rather than prompt history or accumulated examples, carries the learned policy. We demonstrate the method with CatSynth, a synthetic browser application and captured coding-agent experiment. In one open-world run with GPT-5.4-mini, 8 of 14 frozen candidate cases became counterexamples. The rebuild controls inherited that promotion schedule, and all three paths passed the 8 accepted cases. Rebuilding from the evolved sketch passed 19 of 21 withheld cases, compared with 15 of 21 when rebuilding from the initial sketch and replaying all accepted examples. Retaining code across counterexamples required 9 Developer calls and 719 lines of cumulative artifact churn, versus 15 calls and 2,394 lines for replay-all, and passed 18 of 21 withheld cases. These results provide inspectable evidence that the evolved sketch carried reviewed policy and that retaining code reduced rework in this run; with one model and one reveal order, they do not establish general superiority or correctness beyond the encoded checks.

Adversarial threats against LLMs are escalating faster than current defenses can adapt. We expose a critical geometric blind spot in alignment: adversarial prompts exploit latent camouflage, embedding perilously close to the safe representation manifold while encoding unsafe intent thereby evading surface level defenses like Direct Preference Optimization (DPO), which remain blind to the latent geometry. We introduce ALKALI, the first rigorously curated adversarial benchmark and the most comprehensive to date spanning 9,000 prompts across three macro categories, six subtypes, and fifteen attack families. Evaluation of 21 leading LLMs reveals alarmingly high Attack Success Rates (ASRs) across both open and closed source models, exposing an underlying vulnerability we term latent camouflage, a structural blind spot where adversarial completions mimic the latent geometry of safe ones. To mitigate this vulnerability, we introduce GRACE - Geometric Representation Aware Contrastive Enhancement, an alignment framework coupling preference learning with latent space regularization. GRACE enforces two constraints: latent separation between safe and adversarial completions, and adversarial cohesion among unsafe and jailbreak behaviors. These operate over layerwise pooled embeddings guided by a learned attention profile, reshaping internal geometry without modifying the base model, and achieve up to 39% ASR reduction. Moreover, we introduce AVQI, a geometry aware metric that quantifies latent alignment failure via cluster separation and compactness. AVQI reveals when unsafe completions mimic the geometry of safe ones, offering a principled lens into how models internally encode safety. We make the code publicly available at https://anonymous.4open.science/r/alkali-B416/README.md.

Prior work has shown that fine-tuning models on a narrow domain with misaligned data can lead to broad misalignment - a phenomenon termed "emergent misalignment" (Betley et al. 2025). While all tested models were susceptible to emergent misalignment, some models showed more resistance than others. Specifically the Qwen-2.5 family proved to be relatively resistant, while GPT-4o exhibited the strongest misalignment. In this paper we evaluate if current-generation open-weights models exhibit similar resistance to the Qwen-2.5 family and measure misalignment robustness over a range of model architectures and scales. We replicate the effect across nine modern open-weights models (Gemma 3 and Qwen 3 families, 1B-32B parameters). Models fine-tuned on insecure code generation show a 0.68% misalignment rate (compared to 0.07% for base models), matching the lower end of prior open-model results but dramatically lower than GPT-4o's 20%. We identify a critical format-dependent vulnerability: requiring JSON output doubles misalignment rates compared to natural language prompts (0.96% vs 0.42%). This suggests that structural constraints may bypass safety training by reducing the model's 'degrees of freedom' to refuse. These findings confirm emergent misalignment as a reproducible phenomenon in modern open-weights models, with rates substantially lower than observed in proprietary systems.

Tool-augmented language-model agents execute multi-step workflows over external systems, resolving an entity once and then acting on it across subsequent steps. Prior work shows that in single-step actions, agents select the correct tool but bind it to the wrong entity 24-26% of the time. We study what happens to entity bindings over time: do they stay correct, silently drift to a different entity, or, if wrong from the start, propagate and compound? We formalize binding drift (correct at step 1, wrong later) as distinct from error propagation (wrong at step 1, carried forward), and score them on disjoint workflow sets so the two cannot be conflated. In a controlled multi-step testbed (200 workflows, 580 entity-binding-scored steps, four enterprise domains, eight model backends spanning small to frontier), we find: (1) under controlled error injection, an entity lock (the intuitive "persist the first binding" fix) amplifies wrong actions from 907 to 2,746 (3.0x; bootstrap 95% CI [2.8, 3.3]), because it faithfully carries the seeded wrong entity into every later step; (2) the amplification reaches 8.5x on the most affected model (Claude Opus 4.5); (3) a practical LLM-based re-verifier (a single cheap second model call re-reading the original instruction) reduces wrong actions by 79% (0.21x; CI [0.18, 0.25]), closing the gap to within 1 percentage point of an oracle upper-bound (0.20x); and (4) in the natural (non-injected) setting, baseline agents drift on 18% of eligible workflows, with the per-step error rate rising across steps. Persistence and re-verification are not interchangeable: a defense that eliminates drift can worsen propagation, and a practical re-verifier nearly matches oracle recovery.

Emergent misalignment can arise when a language model is fine-tuned on a narrowly scoped supervised objective: the model learns the target behavior, yet also develops undesirable out-of-domain behaviors. We investigate a mechanistic approach to preventing emergent misalignment by identifying a small set of internal features that reliably control the misaligned behavior and then discouraging the model from strengthening these features during fine-tuning. Across six fine-tuning domains, blocking (i.e., constraining) a fixed set of features achieves up to 95\% relative reduction in emergent misalignment with no degradation in model quality or target-task performance. We strengthen validity with disjoint selection/evaluation splits, multiple independent judges, multiple random seeds for key settings, quality metrics, and extensive ablations demonstrating that the reduction in misalignment is specific to the identified mechanism. We also characterize a limiting regime in which misalignment re-emerges under prolonged fine-tuning, present evidence consistent with rerouting through alternative features or layers, and evaluate modifications that partially restore the misalignment-blocking effect. Overall, our results show that targeted training-time constraints on internal mechanisms can mitigate emergent misalignment without degrading target-task performance.

Large language models (LLMs) are reshaping numerous facets of our daily lives, leading widespread adoption as web-based services. Despite their versatility, LLMs face notable challenges, such as generating hallucinated content and lacking access to up-to-date information. Lately, to address such limitations, Retrieval-Augmented Generation (RAG) has emerged as a promising direction by generating responses grounded in external knowledge sources. A typical RAG system consists of i) a retriever that probes a group of relevant passages from a knowledge base and ii) a generator that formulates a response based on the retrieved content. However, as with other AI systems, recent studies demonstrate the vulnerability of RAG, such as knowledge corruption attacks by injecting misleading information. In response, several defense strategies have been proposed, including having LLMs inspect the retrieved passages individually or fine-tuning robust retrievers. While effective, such approaches often come with substantial computational costs. In this work, we introduce RAGDefender, a resource-efficient defense mechanism against knowledge corruption (i.e., by data poisoning) attacks in practical RAG deployments. RAGDefender operates during the post-retrieval phase, leveraging lightweight machine learning techniques to detect and filter out adversarial content without requiring additional model training or inference. Our empirical evaluations show that RAGDefender consistently outperforms existing state-of-the-art defenses across multiple models and adversarial scenarios: e.g., RAGDefender reduces the attack success rate (ASR) against the Gemini model from 0.89 to as low as 0.02, compared to 0.69 for RobustRAG and 0.24 for Discern-and-Answer when adversarial passages outnumber legitimate ones by a factor of four (4x).

While Large Language Models (LLMs) have demonstrated impressive performance in various domains and tasks, concerns about their safety are becoming increasingly severe. In particular, since models may store unsafe knowledge internally, machine unlearning has emerged as a representative paradigm to ensure model safety. Existing approaches employ various training techniques, such as gradient ascent and negative preference optimization, in attempts to eliminate the influence of undesired data on target models. However, these methods merely suppress the activation of undesired data through parametric training without completely eradicating its informational traces within the model. This fundamental limitation makes it difficult to achieve effective continuous unlearning, rendering these methods vulnerable to relearning attacks. To overcome these challenges, we propose a Metamorphosis Representation Projection (MRP) approach that pioneers the application of irreversible projection properties to machine unlearning. By implementing projective transformations in the hidden state space of specific network layers, our method effectively eliminates harmful information while preserving useful knowledge. Experimental results demonstrate that our approach enables effective continuous unlearning and successfully defends against relearning attacks, achieving state-of-the-art performance in unlearning effectiveness while preserving natural performance. Our code is available in https://github.com/ChengcanWu/MRP.

As Large Language Models quickly become ubiquitous, it becomes critical to understand their security vulnerabilities. Recent work shows that text optimizers can produce jailbreaking prompts that bypass moderation and alignment. Drawing from the rich body of work on adversarial machine learning, we approach these attacks with three questions: What threat models are practically useful in this domain? How do baseline defense techniques perform in this new domain? How does LLM security differ from computer vision? We evaluate several baseline defense strategies against leading adversarial attacks on LLMs, discussing the various settings in which each is feasible and effective. Particularly, we look at three types of defenses: detection (perplexity based), input preprocessing (paraphrase and retokenization), and adversarial training. We discuss white-box and gray-box settings and discuss the robustness-performance trade-off for each of the defenses considered. We find that the weakness of existing discrete optimizers for text, combined with the relatively high costs of optimization, makes standard adaptive attacks more challenging for LLMs. Future research will be needed to uncover whether more powerful optimizers can be developed, or whether the strength of filtering and preprocessing defenses is greater in the LLMs domain than it has been in computer vision.

Large Language Models (LLMs) are increasingly used in a variety of important applications, yet their safety and reliability remain as major concerns. Various adversarial and jailbreak attacks have been proposed to bypass the safety alignment and cause the model to produce harmful responses. We introduce Self-supervised Prompt INjection (SPIN) which can detect and reverse these various attacks on LLMs. As our self-supervised prompt defense is done at inference-time, it is also compatible with existing alignment and adds an additional layer of safety for defense. Our benchmarks demonstrate that our system can reduce the attack success rate by up to 87.9%, while maintaining the performance on benign user requests. In addition, we discuss the situation of an adaptive attacker and show that our method is still resilient against attackers who are aware of our defense.

Despite significant progress in alignment, large language models (LLMs) remain vulnerable to adversarial attacks that elicit harmful behaviors. Activation steering techniques offer a promising inference-time intervention approach, but existing methods suffer from critical limitations: activation addition requires careful coefficient tuning and is sensitive to layer-specific norm variations, while directional ablation provides only binary control. Recent work on Angular Steering introduces continuous control via rotation in a 2D subspace, but its practical implementation violates norm preservation, causing distribution shift and generation collapse, particularly in models below 7B parameters. We propose Selective Steering, which addresses these limitations through two key innovations: (1) a mathematically rigorous norm-preserving rotation formulation that maintains activation distribution integrity, and (2) discriminative layer selection that applies steering only where feature representations exhibit opposite-signed class alignment. Experiments across nine models demonstrate that Selective Steering achieves 5.5x higher attack success rates than prior methods while maintaining zero perplexity violations and approximately 100\% capability retention on standard benchmarks. Our approach provides a principled, efficient framework for controllable and stable LLM behavior modification. Code: https://github.com/knoveleng/steering

Small language models (SLMs) are increasingly deployed on edge devices, making their safety alignment crucial yet challenging. Current shallow alignment methods that rely on direct refusal of malicious queries fail to provide robust protection, particularly against adversarial jailbreaks. While deliberative safety reasoning alignment offers deeper alignment for defending against sophisticated attacks, effectively implanting such reasoning capability in SLMs with limited capabilities remains an open challenge. Moreover, safety reasoning incurs significant computational overhead as models apply reasoning to nearly all queries, making it impractical for resource-constrained edge deployment scenarios that demand rapid responses. We propose EASE, a novel framework that enables practical and Efficient safety Alignment for Small languagE models. Our approach first identifies the optimal safety reasoning teacher that can effectively distill safety reasoning capabilities to SLMs. We then align models to selectively activate safety reasoning for dangerous adversarial jailbreak queries while providing direct responses to straightforward malicious queries and general helpful tasks. This selective mechanism enables small models to maintain robust safety guarantees against sophisticated attacks while preserving computational efficiency for benign interactions. Experimental results demonstrate that EASE reduces jailbreak attack success rates by up to 17% compared to shallow alignment methods while reducing inference overhead by up to 90% compared to deliberative safety reasoning alignment, making it practical for SLMs real-world edge deployments.

Recent advances in large reasoning models (LRMs) have enabled remarkable performance on complex tasks such as mathematics and coding by generating long Chain-of-Thought (CoT) traces. In this paper, we identify and systematically analyze a critical vulnerability we term reasoning distraction, where LRMs are diverted from their primary objective by irrelevant yet complex tasks maliciously embedded in the prompt. Through a comprehensive study across diverse models and benchmarks, we show that even state-of-the-art LRMs are highly susceptible, with injected distractors reducing task accuracy by up to 60%. We further reveal that certain alignment techniques can amplify this weakness and that models may exhibit covert compliance, following hidden adversarial instructions in reasoning while concealing them in the final output. To mitigate these risks, we propose a training-based defense that combines Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL) on synthetic adversarial data, improving robustness by over 50 points on challenging distractor attacks. Our findings establish reasoning distraction as a distinct and urgent threat to LRM reliability and provide a practical step toward safer and more trustworthy reasoning systems.

Reinforcement Learning with Verifiable Rewards (RLVR) is increasingly viewed as a tree pruning mechanism. However, we identify a systemic pathology termed Recursive Space Contraction (RSC), an irreversible collapse driven by the combined dynamics of positive sharpening and negative squeezing, where the sampling probability of valid alternatives vanishes. While Kullback-Leibler (KL) regularization aims to mitigate this, it imposes a rigid Shape Matching constraint that forces the policy to mimic the reference model's full density, creating a gradient conflict with the sharpening required for correctness. We propose Anchored Policy Optimization (APO), shifting the paradigm from global Shape Matching to Support Coverage. By defining a Safe Manifold based on the reference model's high-confidence support, APO permits aggressive sharpening for efficiency while selectively invoking a restorative force during error correction to prevent collapse. We theoretically derive that APO serves as a gradient-aligned mechanism to maximize support coverage, enabling an Elastic Recovery that re-inflates valid branches. Empirical evaluations on mathematical benchmarks demonstrate that APO breaks the accuracy-diversity trade-off, significantly improving Pass@1 while restoring the Pass@K diversity typically lost by standard policy gradient methods.

Harmful fine-tuning (HFT), performed directly on open-source LLMs or through Fine-tuning-as-a-Service, breaks safety alignment and poses significant threats. Existing methods aim to mitigate HFT risks by learning robust representation on alignment data or making harmful data unlearnable, but they treat each data sample equally, leaving data vulnerability patterns understudied. In this work, we reveal that certain subsets of alignment data are consistently more prone to forgetting during HFT across different fine-tuning tasks. Inspired by these findings, we propose Vulnerability-Aware Alignment (VAA), which estimates data vulnerability, partitions data into "vulnerable" and "invulnerable" groups, and encourages balanced learning using a group distributionally robust optimization (Group DRO) framework. Specifically, VAA learns an adversarial sampler that samples examples from the currently underperforming group and then applies group-dependent adversarial perturbations to the data during training, aiming to encourage a balanced learning process across groups. Experiments across four fine-tuning tasks demonstrate that VAA significantly reduces harmful scores while preserving downstream task performance, outperforming state-of-the-art baselines.