LLM Safety Alignment

LLM: Large Language Model

Follow this topic Ask about this topic

Momentum

25 papers in the last four weeks, up 14% on the four weeks before. 0.2% of all new papers.

Jul 13Week of Sep 28

Latest papers 236

Apr 18, 2026cs.CL

When Choices Become Risks: Safety Failures of Large Language Models under Multiple-Choice Constraints

We identify and systematically characterize a class of task-structural alignment failures in large language models (LLMs): even when the harmful intent remains unchanged, changing the task presentation and output constraints can substantially alter model safety behavior. Specifically, when a harmful request is reformulated as a forced-choice multiple-choice question (MCQ) in which all options are harmful and no refusal option is provided, some models that refuse the equivalent open-ended query instead select, prefer, or justify a harmful option. We evaluate 14 proprietary and open-source models on a bilingual Chinese-English human-authored dataset covering five harm categories, together with 900 model-generated Chinese adversarial MCQs. On human-authored data, attack success rate (ASR) increases sharply as prompts shift from open-ended queries to explicit forced-choice formats, typically peaking under intermediate levels of choice constraint. Model-generated Chinese MCQs further weaken or eliminate the recovery regime observed on human-authored data, driving ASR close to saturation for multiple models. The observed transfer patterns are consistent with stronger generators producing more difficult or boundary-adjacent MCQs, although other properties of the generated inputs may also contribute. We also find that adding an explicit refusal option or a safety preamble substantially reduces ASR for several high-capability models, often to near-zero levels, although their effectiveness varies across target models. These findings suggest that safety evaluations centered on open-ended generation may underestimate risks in structured deployment settings, and that task structure should be treated as an important and diagnosable dimension of safety evaluation and alignment training.

Yuheng Chen, Zhiyu Wu, Bowen Cheng +2

Kagoshima University · Fudan University · China University of Petroleum-Beijing

Adversarial Attacks on LLMs LLM Safety Alignment LLM Safety Evaluation Refusal Behavior in Language Models

Apr 18, 2026cs.CL

DART: Mitigating Harm Drift in Difference-Aware LLMs via Distill-Audit-Repair Training

Large language models (LLMs) tuned for safety often avoid acknowledging demographic differences, even when such acknowledgment is factually correct (e.g., ancestry-based disease incidence) or contextually justified (e.g., religious hiring preferences). This identity-blindness yields incorrect responses, unnecessary refusals, or generic "equal-treatment" defaults. We study this via difference-awareness classification: given a question involving demographic groups, the task is not to answer directly, but to classify whether a correct answer requires recognizing group differences (yes) or whether groups should be treated identically (no). Crucially, fine-tuning for accuracy triggers harm drift: model-generated explanations become increasingly harmful as decision accuracy improves, whether by elaborating harmful content, introducing problematic assumptions, or failing to flag harms the baseline identified. To mitigate this, we introduce DART (Distill--Audit--Repair Training), which distills label-conditioned reasoning from a teacher, audits outputs for harm drift cases relative to baseline, and repairs problematic cases via severity-weighted fine-tuning. On eight benchmarks, DART improves Llama-3-8B-Instruct accuracy from 39.0% to 68.8%, with largest gains on equal-treatment prompts (11.3% -> 72.6%), while reducing harm drift cases by 72.6%. It also transfers to 280 open-ended real-world queries across medical, legal, policy, and educational domains, improving difference-appropriate responses from 39.8% to 77.5% while reducing refusals from 34.3% to 3.0%. Our results demonstrate that accuracy and safety need not conflict when explicit detection and repair mechanisms are in place.

Ziwen Pan, Zihan Liang, Jad Kabbara +1

Emory University · MIT

LLM Auditing Social Bias in Language Models LLM Safety Alignment LLM Safety

Apr 17, 2026cs.CR

Benign Fine-Tuning Breaks Safety Alignment in Audio LLMs

Fine-tuning on benign data is known to degrade safety alignment in text and vision LLMs, but whether distinct input properties drive this vulnerability differently remains unclear. Audio introduces a richer problem where benign samples can neighbor harmful content through what is said or how it sounds. We present the first systematic study of benign fine-tuning safety in Audio LLMs, evaluating three state-of-the-art models with a proximity-based framework that decomposes embedding-space distance into semantic, acoustic, and mixed axes. We find that the dominant vulnerability axis is architecture-conditioned, determined by how each model's encoder and projector transform audio into the backbone LLM's input space. Across three models, benign fine-tuning elevates Jailbreak Success Rate (JSR) from single digits to as high as 87%, with the most damaging axis shifting from semantic to acoustic proximity depending on encoder design. Mechanistically, fine-tuning selectively suppresses late-layer refusal circuits while frozen encoders preserve upstream representations: the model still detects harmful content but stops refusing, a recognition-refusal dissociation. Two practical defenses, filtering training data to maximize distance from harmful embeddings and a textual system prompt at inference, reduce JSR to near-zero without architectural modification. These findings show that safety evaluation should account for modality and architecture, while highlighting Audio LLMs as a useful testbed for understanding alignment fragility.

Jaechul Roh, Virat Shejwalkar, Amir Houmansadr

University of Massachusetts Amherst

Language Model Safety Evaluation LLM Fine-Tuning LLM Safety Alignment Audio-Language Models LLM Safety

Apr 17, 2026cs.CY

From Consumption to Reflection: Designing Human-AI Relations for Stable Reasoning

Large language models (LLMs) have transformed how humans access information, but not how we reason with it. Their fluency accelerates consumption while bypassing the slow, reflective processes that underpin sound judgment. This paper introduces Relational Reflective Intelligence (RRI), an inference-time governance layer that operationalizes reflection through auditable reasoning loops. RRI operates not inside the model but around it, providing a practical structure for stable, auditable reasoning between humans and LLMs. The core premise is that LLMs inherit cognitive vulnerabilities similar to those that shape human thought: reliance on intuitive shortcuts, confusion between representation and reality, and a preference for coherence over falsification. When humans and models share these tendencies, their errors compound. We refer to this as relational drift, a failure that arises from interaction rather than from the model alone. Addressing this requires a shift from modeling relations between words to structuring relations between model outputs and human reasoning. RRI provides this missing layer through three components: the Rose-Frame, which identifies likely breakdowns in reasoning; the Architect's Pen, which introduces targeted reflection steps at critical moments; and an inference-time workflow that embeds these steps without retraining the model. Together, these elements transform human-AI interaction into a joint reasoning system with explicit checkpoints, conflict surfacing, and an auditable trail of assumptions. Rather than making machines think like humans or forcing humans to reason like machines, RRI creates a structured interaction in which both compensate for each other's limitations. It reframes AI safety as a cognitive architecture problem, where reliable decisions depend on embedding reflection directly into the interaction process.

Rikard Rosenbacke, Carl Rosenbacke, Victor Rosenbacke +1

Faculty of Medicine, Lund University, Sweden · Department of Economics, Lund University School of Economics and Management, Sweden · Department of Health Services Research and Policy, London School of Hygiene & Tropical Medicine, UK

LLM Safety Alignment Human Oversight of AI Human-AI Collaboration

Apr 17, 2026cs.LG

Pruning Unsafe Tickets: A Resource-Efficient Framework for Safer and More Robust LLMs

Machine learning models are increasingly deployed in real-world applications, but even aligned models such as Mistral and LLaVA still exhibit unsafe behaviors inherited from pre-training. Current alignment methods like SFT and RLHF primarily encourage models to generate preferred responses, but do not explicitly remove the unsafe subnetworks that trigger harmful outputs. In this work, we introduce a resource-efficient pruning framework that directly identifies and removes parameters associated with unsafe behaviors while preserving model utility. Our method employs a gradient-free attribution mechanism, requiring only modest GPU resources, and generalizes across architectures and quantized variants. Empirical evaluations on ML models show substantial reductions in unsafe generations and improved robustness against jailbreak attacks, with minimal utility loss. From the perspective of the Lottery Ticket Hypothesis, our results suggest that ML models contain "unsafe tickets" responsible for harmful behaviors, and pruning reveals "safety tickets" that maintain performance while aligning outputs. This provides a lightweight, post-hoc alignment strategy suitable for deployment in resource-constrained settings.

Wai Man Si, Mingjie Li, Michael Backes +1

CISPA Helmholtz Center for Information Security

LLM Pruning LLM Safety Alignment LLM Safety Neural Network Pruning

Apr 14, 2026cs.LG

Safety Training Modulates Harmful Misalignment Under On-Policy RL, But Direction Depends on Environment Design

Specification gaming under Reinforcement Learning (RL) is known to cause LLMs to develop sycophantic, manipulative, or deceptive behavior, yet the conditions under which this occurs remain unclear. We train 11 instruction-tuned LLMs (0.5B-14B) with on-policy RL across 3 environments and find that model size acts as a safety buffer in some environments but enables greater harmful exploitation in others. Controlled ablations trace this reversal to environment-specific features such as role framing and implicit gameability cues. We further show that most safety benchmarks do not predict RL-induced misalignment, except in the case of Sycophancy scores when the exploit relies on inferring the user's preference. Finally, we find that on-policy RL preserves a safety buffer inherent in the model's own generation distribution, one that is bypassed during off-policy settings.

Leon Eshuijs, Shihan Wang, Antske Fokkens

Vrije Universiteit Amsterdam · Utrecht University

Reward Hacking RL for Language Models LLM Safety Alignment LLM Safety Evaluation

Apr 9, 2026cs.AI

Activation Steering for Aligned Open-ended Generation without Sacrificing Coherence

Alignment in LLMs is more brittle than commonly assumed: misalignment can be induced 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 activation steering, which could be used as a lightweight runtime defense. We implement three methods: Steer-With-Fixed-Coefficient (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 the threshold. We evaluate these methods on two threat models, dishonesty and dismissiveness, using malicious system prompts as a controlled proxy for misalignment. We conduct our experiments on two architectures (Llama-3.3-70B-Instruct and Qwen3.6-27B). All methods substantially recover alignment. StTP and StMP preserve general capabilities (MMLU, MT-Bench, AlpacaEval) better than uniform steering. Finally, we show that our honesty steering generalizes to out-of-distribution scenarios: a single honesty direction extracted from the aligned model significantly raises scores on the MASK benchmark, suppresses deception in multi-agent settings (Among Us), doubles the hidden-behavior discovery rate on AuditBench, and restores honesty in an emergently misaligned model

Niklas Herbster, Martin Zborowski, Alberto Tosato +2

1Tara Research · 2Technical University of Munich · 3Mila Quebec AI Institute

LLM Safety Alignment Language Model Robustness Activation Steering LLM Honesty

Mar 26, 2026cs.CL

SafeMath: Safe Solutions for Unsafe Math Word Problems

Recent research points toward LLMs being manipulated through adversarial and seemingly benign inputs, resulting in harmful, biased, or policy-violating outputs. In this paper, we study an underexplored issue concerning harmful and toxic mathematical word problems. We show that math questions, particularly those framed as natural language narratives, can serve as a subtle medium for propagating biased, unethical, or psychologically harmful content, with heightened risks in educational settings involving children. To support a systematic study of this phenomenon, we introduce ToxicGSM, a dataset of 1.9k arithmetic problems in which harmful or sensitive context is embedded while preserving mathematically well-defined reasoning tasks. Using this dataset, we audit the behaviour of existing LLMs and analyse the trade-offs between safety enforcement and mathematical correctness. We further propose SafeMath -- a safety alignment technique that reduces harmful outputs while maintaining, and in some cases improving, mathematical reasoning performance. Our results highlight the importance of disentangling linguistic harm from math reasoning and demonstrate that effective safety alignment need not come at the cost of accuracy.

Sagnik Basu, Subhrajit Mitra, Aman Juneja +3

1Indian Institute of Technology Kharagpur · 2Cisco Systems · 3National University of Singapore

LLM Safety Alignment LLM Safety Evaluation LLM Reasoning

Mar 11, 2026cs.LG

Safe RLHF Beyond Expectation: Stochastic Dominance for Universal Spectral Risk Control

Safe Reinforcement Learning from Human Feedback (RLHF) typically enforces safety through expected cost constraints, but the expectation captures only a single statistic of the cost distribution and fails to account for distributional uncertainty, particularly under heavy tails or rare catastrophic events. This limitation is problematic when robustness and risk sensitivity are critical. Stochastic dominance offers a principled alternative by comparing entire cost distributions rather than just their averages, enabling direct control over tail risks and potential out-of-distribution failures that expectation-based constraints may overlook. In this work, we propose Risk-sensitive Alignment via Dominance (RAD), a novel alignment framework that replaces scalar expected cost constraints with First-Order Stochastic Dominance (FSD) constraints. We operationalize this constraint by comparing the target policy's cost distribution to that of a reference policy within an Optimal Transport (OT) framework, using entropic regularization and Sinkhorn iterations to obtain a differentiable and computationally efficient objective for stable end-to-end optimization. Furthermore, we introduce quantile-weighted FSD constraints and show that weighted FSD universally controls a broad class of Spectral Risk Measures (SRMs), so that improvements under weighted dominance imply guaranteed improvements in the corresponding spectral risk. This provides a principled mechanism for tuning a model's risk profile via the quantile weighting function. Empirical results demonstrate that RAD improves harmlessness over baselines while remaining competitive in helpfulness, and exhibits greater robustness on out-of-distribution harmlessness evaluations.

Yaswanth Chittepu, Ativ Joshi, Rajarshi Bhattacharjee +1

University of Massachusetts Amherst

Constrained RL LLM Safety Alignment Risk-Sensitive RL Decision-Making under Uncertainty

Feb 14, 2026cs.CR

Mitigating the Safety-utility Trade-off in LLM Alignment via Adaptive Safe Context Learning

While reasoning models have achieved remarkable success in complex reasoning tasks, their increasing power necessitates stringent safety measures. For safety alignment, the core challenge lies in the inherent trade-off between safety and utility. However, prevailing alignment strategies typically construct CoT training data with explicit safety rules via context distillation. This approach inadvertently limits reasoning capabilities by creating a rigid association between rule memorization and refusal. To mitigate the safety-utility trade-off, we propose the Adaptive Safe Context Learning~(ASCL) framework to improve the reasoning given proper context. ASCL formulates safety alignment as a multi-turn tool-use process, empowering the model to autonomously decide when to consult safety rules and how to generate the ongoing reasoning. Furthermore, to counteract the preference for rule consultation during RL, we introduce Inverse Frequency Policy Optimization~(IFPO) to rebalance advantage estimates. By decoupling rule retrieval and subsequent reasoning, our method achieves higher overall performance compared to baselines. Our code is publicly available at https://github.com/ybwang119/ASCL.

Yanbo Wang, Minzheng Wang, Jian Liang +3

School of Artificial Intelligence, University of Chinese Academy of Sciences, Beijing, China. · NLPR & MAIS, Institute of Automation, Chinese Academy of Sciences, Beijing, China. · Ritzz-AI.

LLM Alignment RL for Language Model Reasoning LLM Safety Alignment Multi-Objective Language Model Alignment

Jan 8, 2026cs.CL

Towards Multi-modal Multi-turn Safety: From Agentic Interaction to Strategic Alignment

Despite remarkable capability in multi-modal understanding, deploying Multi-modal Large Language Models (MLLMs) in open-ended conversational scenarios introduces safety risks that remain poorly addressed by existing alignment methods. Unlike simple malicious visual question and answer (VQA) pairs , multi-turn interactions enable adversaries to incrementally reconstruct harmful intent across dialogues, progressively bypassing safety constraints in ways that are difficult to detect at any individual turn. Meanwhile, conventional reinforcement learning from human feedback (RLHF) approaches are unsuitable for this situation: designed primarily for VQA tasks, they neither capture cross-turn risk dynamics nor scale efficiently without costly manual preference annotation. To close this gap, we introduce \textbf{MINT-Safe}, an open-source visual multi-turn training dataset comprising 11,270 multi-image dialogues and 500 refusal VQA pairs, constructed via multi-agent interaction with text-to-image (T2I) tool-call augmentation. Building on MINT-Safe, we propose \textbf{TAD-Align}, a dialogue safety alignment framework centered on a turn-aware dual-objective reward function. Rather than treating all dialogue turns uniformly, TAD-Align leverages rollout-based safety score variance to dynamically identify turns where the model exhibits inconsistent safety behavior, and adaptively up-weights these turns during optimization. Experiments on Qwen2.5-VL-7B-Instruct and LLaVA-NeXT-7B demonstrate reductions of over 10% in Attack Success Rate (ASR), alongside improvements of at least 8% in harmlessness and 13% in helpfulness on multi-modal multi-turn safety benchmarks, while preserving general model capabilities.

Han Zhu, Jiale Chen, Chengkun Cai +8

Hong Kong University of Science and Technology · University College London · AISpeech +1

Language Model Safety Evaluation Multimodal Large Language Models LLM Safety Alignment Multimodal Model Evaluation

Dec 18, 2025cs.CR

Agent Tools Orchestration Leaks More: Dataset, Benchmark, and Mitigation

LLM agents can combine individually non-revealing tool returns and disclose a sensitive conclusion, creating Tools Orchestration Privacy Risk (TOP-R). We formalize TOP-R through three conditions: conclusion sensitivity, single-source non-inferability, and compositional inferability. We introduce Library-Grounded Reverse-Inference Seed Expansion (LRSE), a four-library reverse-construction pipeline, and use it to build TOP-Bench, a 1,000-instance benchmark evaluated under a controlled two-stage tool-use protocol. Across six LLM agents, average task completion, leakage, and H-score are 98.0 percent, 88.6 percent, and 20.4. With native reasoning enabled, four models average 81.4 percent final-response leakage and 82.4 percent reasoning-trace leakage. With reasoning disabled, three prompt-only safeguards improve H-score by an average of about 3.4 points on TOP-Bench. We further propose TOP-Align, an SFT+DPO method for learning safer task-completion boundaries. On a separate post-training evaluation set, TOP-Align improves H-score by 16.2 points over the base model, versus a 5.0-point average gain from prompt-only mitigation on the same set. These results show that TOP-R requires defenses beyond prompting alone. Dataset and code are available at https://github.com/1Ponder/TOP-R.

Yuxuan Qiao, Dongqin Liu, Hongchang Yang +2

Institute of Information Engineering, Chinese Academy of Sciences · School of Cyber Security, University of Chinese Academy of Sciences

LLM Agent Security LLM Agents LLM Safety Alignment

Dec 8, 2025cs.CL

Response-Based Knowledge Distillation for Multilingual Jailbreak Prevention Unwittingly Compromises Safety

Large language models (LLMs) are increasingly deployed worldwide, yet their safety alignment remains predominantly English-centric. This allows for vulnerabilities in non-English contexts, especially with low-resource languages. We introduce a novel application of knowledge distillation (KD) in the context of multilingual jailbreak prevention, examining its efficacy. We distill the refusal behaviors of a proprietary teacher model (OpenAI o1-mini) with Low-Rank Adaptation (LoRA) into three open-source student models: Meta-Llama-3-8B-Instruct, Gemma-2-2B-IT, and Qwen3-8B, using ~28,000 multilingual jailbreak prompts from XSafety via black-box response-based, parameter-efficient fine-tuning (PEFT). Evaluation on the MultiJail benchmark reveals a counterintuitive behavior: standard fine-tuning on the teacher's ``safe'' refusal data inadvertently increases Jailbreak Success Rate (JSR) for all student models, up to 16.6 percentage points. Our experiments reveal a divergent generalization to unseen languages during distillation, with varying outcomes depending on the base model. By removing a primary source of safety degradation, nuanced `boundary' refusals, we mitigate or even reverse safety declines in student models, although reductions in reasoning performance (GSM8K) persist. Overall, our exploratory study highlights the challenges and potential of KD as a technique for multilingual safety alignment, offering a foundation for future research in this direction.

Max Zhang, Derek Liu, Kai Zhang +2

Adversarial Attacks on LLMs LLM Safety Alignment Jailbreak Defense Knowledge Distillation

Oct 30, 2025cs.CL

Reasoning Up the Instruction Ladder for Controllable Language Models

As large language model (LLM) based systems take on high-stakes roles in real-world decision-making, they must reconcile competing instructions from multiple sources within a single prompt context. Enforcing an instruction hierarchy, where higher-level directives override lower-priority requests, is critical to the reliability and control of LLMs. In this work, we reframe instruction hierarchy resolution as a reasoning task. The model must first "think" about the relationship between a given user prompt and higher-priority instructions before generating a response. To enable this capability, we construct VerIH, a training dataset of constraint-following tasks with verifiable answers, comprising aligned and conflicting system-user instructions. We show that lightweight reinforcement learning with VerIH effectively transfers general reasoning capabilities of models to instruction prioritization. Our method leads to consistent improvements across multiple model families on both instruction following and instruction hierarchy benchmarks, achieving ~20% absolute improvement in conflict setups. Our method also leads to improved alignment to safety-critical scenarios beyond the training distribution, exhibiting increased robustness against jailbreak and prompt injection, reducing absolute attack success rates by up to 20%. Our results establish reasoning over instruction hierarchies as a practical mechanism for improving AI reliability, where targeted updates to system prompts produce predictable, controllable, and robust changes in model behavior.

Zishuo Zheng, Vidhisha Balachandran, Chan Young Park +2

Department of Computer Science and Engineering, The Ohio State University · Microsoft Research · Allen Institute for AI

LLM Safety Alignment Instruction Following LLM Reliability

Oct 15, 2025cs.CV

Attention Misses Visual Risk: Risk-Adaptive Steering for Multimodal Safety Alignment

Even modern AI models often remain vulnerable to multimodal queries in which harmful intent is embedded in images. A widely used approach for safety alignment is training with extensive multimodal safety datasets, but the costs of data curation and training are often prohibitive. To mitigate these costs, inference-time alignment has recently been explored, but they often lack generalizability across diverse multimodal jailbreaks and still incur notable overhead due to extra forward passes for response refinement or heavy pre-deployment calibration procedures. Here, we identify insufficient visual attention to safety-critical image regions as one of the key causes of multimodal safety failures. Building on this insight, we propose Multimodal Risk-Adaptive Steering (MoRAS), which enhances safety-critical visual attention via concise visual contexts for accurate multimodal risk assessment. This risk signal enables risk-adaptive steering for direct refusals, reducing inference overhead while remaining generalizable across diverse multimodal jailbreaks. Notably, MoRAS requires only a small calibration set to estimate multimodal risk, substantially reducing pre-deployment overhead. We conduct various empirical validations across multiple benchmarks and MLLM backbones, and observe that the proposed MoRAS consistently mitigates jailbreaks, preserves utility, and reduces computational overhead compared to state-of-the-art inference-time defenses.

Jonghyun Park, Minhyuk Seo, Chaewon Yeo +1

Seoul National University · KU Leuven

Multimodal Robustness LLM Safety Alignment Jailbreak Defense

Oct 7, 2025cs.LG

The Alignment Auditor: A Bayesian Framework for Verifying and Refining LLM Objectives

The objectives that Large Language Models (LLMs) implicitly optimize remain dangerously opaque, making trustworthy alignment and auditing a grand challenge. While Inverse Reinforcement Learning (IRL) can infer reward functions from behaviour, existing approaches either produce a single, overconfident reward estimate or fail to address the fundamental ambiguity of the task (non-identifiability). This paper introduces a principled auditing framework that re-frames reward inference from a simple estimation task to a comprehensive process for verification. Our framework leverages Bayesian IRL to not only recover a distribution over objectives but to enable three critical audit capabilities: (i) Quantifying and systematically reducing non-identifiability by demonstrating posterior contraction over sequential rounds of evidence; (ii) Providing actionable, uncertainty-aware diagnostics that expose spurious shortcuts and identify out-of-distribution prompts where the inferred objective cannot be trusted; and (iii) Validating policy-level utility by showing that the refined, low-uncertainty reward can be used directly in RLHF to achieve training dynamics and toxicity reductions comparable to the ground-truth alignment process. Empirically, our framework successfully audits a detoxified LLM, yielding a well-calibrated and interpretable objective that strengthens alignment guarantees. Overall, this work provides a practical toolkit for auditors, safety teams, and regulators to verify what LLMs are truly trying to achieve, moving us toward more trustworthy and accountable AI.

Matthieu Bou, Nyal Patel, Arjun Jagota +2

1Imperial College London · 2Amazon AGI

Reinforcement Learning LLM Auditing LLM Safety Alignment

Aug 28, 2025cs.CL

Turning the Spell Around: Lightweight Alignment Amplification via Rank-One Safety Injection

Safety alignment in Large Language Models (LLMs) often involves mediating internal representations to refuse harmful requests. Recent research has demonstrated that these safety mechanisms can be bypassed by ablating or removing specific representational directions within the model. In this paper, we propose the opposite approach: Rank-One Safety Injection (ROSI), a white-box method that amplifies a model's safety alignment by permanently steering its activations toward the refusal-mediating subspace. ROSI operates as a simple, fine-tuning-free rank-one weight modification applied to all residual stream write matrices. The required safety direction can be computed from a small set of harmful and harmless instruction pairs. We show that ROSI consistently increases safety refusal rates - as evaluated by Llama Guard 3 - while preserving the utility of the model on standard benchmarks such as MMLU, HellaSwag, and Arc. Furthermore, we show that ROSI can also re-align 'uncensored' models by amplifying their own latent safety directions, demonstrating its utility as an effective last-mile safety procedure. Our results suggest that targeted, interpretable weight steering is a cheap and potent mechanism to improve LLM safety, complementing more resource-intensive fine-tuning paradigms.

Harethah Abu Shairah, Hasan Abed Al Kader Hammoud, George Turkiyyah +1

King Abdullah University of Science and Technology (KAUST)

LLM Guardrails LLM Safety Alignment Model Editing Activation Steering

Aug 9, 2025cs.CR

Context Misleads LLMs: The Role of Context Filtering in Maintaining Safe Alignment of LLMs

While Large Language Models (LLMs) have shown significant advancements in performance, various jailbreak attacks have posed growing safety and ethical risks. Malicious users often exploit adversarial context to deceive LLMs, prompting them to generate responses to harmful queries. In this study, we propose a new defense mechanism called Context Filtering, an input pre-processing method designed to filter out untrustworthy and unreliable context while identifying the primary prompts containing the real user intent to uncover concealed malicious intent. Given that enhancing the safety of LLMs often compromises their helpfulness, potentially affecting the experience of benign users, our method aims to improve the safety of the LLMs while preserving their original performance. We evaluate the effectiveness of our model in defending against jailbreak attacks through comparative analysis, comparing our approach with state-of-the-art defense mechanisms against six different attacks and assessing the helpfulness of LLMs under these defenses. Our model demonstrates its ability to reduce the Attack Success Rates of jailbreak attacks by up to 92% while maintaining the original LLMs' performance, achieving state-of-the-art Safety and Helpfulness balance. Notably, Context Filtering is a plug-and-play method that can be applied to all LLMs, including both white-box and black-box models, to enhance their safety without requiring any fine-tuning of the models themselves. Our model is available for research purposes.

Jinhwa Kim, Ian G. Harris

Department of Computer Science University of California, Irvine

LLM Safety Alignment LLM Safety LLM Jailbreak Attacks

Jun 9, 2025cs.LG

Chasing Moving Targets with Online Self-Play Reinforcement Learning for Safer Language Models

Conventional large language model (LLM) safety alignment relies on a reactive, disjoint loop: attackers exploit a static model, then defenders patch exposed vulnerabilities. This sequential setup leads to attackers overfitting obsolete exploits while defenders perpetually lag behind emerging threats. To address this, we introduce Self-RedTeam, the first fully online self-play multi-agent reinforcement learning (MARL) algorithm that continuously co-evolves attacker and defender for robust safety alignment. A single policy self-plays as both attacker and defender, generating adversarial prompts and defending against them, with a reward model adjudicating outcomes. Each role uses hidden chain-of-thought for strategic planning. Grounded in two-player zero-sum game theory, we establish a theoretical safety guarantee: if the game converges to Nash Equilibrium, the defender produces safe responses against any adversarial input. Empirically, Self-RedTeam generalizes across five models from the Llama and Qwen families, uncovering more diverse attacks (+17.80% SBERT) and improving safety of RLHF-trained models by up to 95% across 14 benchmarks. Our work motivates a shift from reactive patching to proactive co-evolution, enabling LLM safety self-improvement via online self-play MARL. Link to code: https://github.com/mickelliu/selfplay-redteaming

Mickel Liu, Liwei Jiang, Yancheng Liang +4

Paul G. Allen School of Computer Science & Engineering, University of Washington, Seattle, WA, USA · Department of Computer Science, Stanford University, Stanford, CA, USA

Self-Play RL Adversarial Attacks on LLMs LLM Safety Alignment Multi-Agent Reinforcement Learning

May 12, 2025cs.CR

Decoding One Safety Trigger Token for Balancing Safety and Usability in Large Language Models

Large Language Models (LLMs) have been extensively used across diverse domains, including virtual assistants, automated code generation, and scientific research. However, they remain vulnerable to jailbreak attacks, which manipulate the models into generating harmful responses despite safety alignment. Recent studies have shown that current safety-aligned LLMs undergo shallow safety alignment. In this work, we conduct an in-depth investigation into the underlying mechanism of this phenomenon and reveal that it manifests through learned ''safety trigger tokens'' that activate the model's safety patterns when paired with the specific input. Through both analysis and empirical verification, we further demonstrate the high similarity of the safety trigger tokens across different harmful inputs. Accordingly, we propose D-STT, a simple yet effective defense algorithm that identifies and explicitly decodes safety trigger tokens of the given safety-aligned LLM to activate the model's learned safety patterns. In this process, the safety trigger is constrained to a single token, which effectively preserves model usability by introducing minimum intervention in the decoding process. Extensive experiments across diverse jailbreak attacks and benign prompts demonstrate that D-STT significantly reduces output harmfulness while preserving model usability and incurring negligible response time overhead, outperforming ten baseline methods.

Haoran Gu, Handing Wang, Yi Mei +2

Xidian University · Victoria University of Wellington · Westlake University

LLM Safety Alignment LLM Safety LLM Jailbreak Attacks Jailbreak Defense

Apr 27, 2025cs.LG

Adaptive Helpfulness-Harmlessness Alignment with Preference Vectors

Ensuring that large language models (LLMs) are both helpful and harmless is a critical challenge, as overly strict constraints can lead to excessive refusals, while permissive models risk generating harmful content. Existing approaches, such as reinforcement learning from human feedback (RLHF) and direct preference optimization (DPO), attempt to balance these trade-offs but suffer from performance conflicts, limited controllability, and poor extendability. To address these issues, we propose Preference Vector, a novel framework inspired by task arithmetic. Instead of optimizing multiple preferences within a single objective, we train separate models on individual preferences, extract behavior shifts as preference vectors, and dynamically merge them at test time. This modular approach enables fine-grained, user-controllable preference adjustments and facilitates seamless integration of new preferences without retraining. Experiments show that our proposed Preference Vector framework improves helpfulness without excessive conservatism, allows smooth control over preference trade-offs, and supports scalable multi-preference alignment.

Ren-Wei Liang, Chin-Ting Hsu, Chan-Hung Yu +6

National Taiwan University · Appier AI Research · Graduate Institute of Communication Engineering, National Taiwan University +1

LLM Alignment LLM Safety Alignment Multi-Objective Language Model Alignment

Jun 3, 2024cs.CL

Decoupled Alignment for Robust Plug-and-Play Adaptation

We introduce a training-free safety enhancement method for aligning large language models (LLMs) without the need for supervised fine-tuning or reinforcement learning from human feedback. Our main idea is to provide a robust plug-and-play approach to prevent shadow alignment when models are adapted to downstream tasks. Specifically, we leverage knowledge distillation to extract alignment signals from well-aligned LLMs and inject them into shadow-aligned models via model fusion, enabling plug-and-play alignment correction. In our methodology, we employ delta debugging to identify the critical components of knowledge necessary for effective distillation. On the harmful question dataset, our method significantly enhances the average defense success rate by approximately 14.42%, reaching as high as 51.39% across 17 influenced LLMs, without compromising performance. Our code is available at https://github.com/NWULIST/DAPA.

Haozheng Luo, Jiahao Yu, Wenxin Zhang +9

Northwestern University · New York University Abu Dhabi · Stanford University +3

LLM Alignment LLM Safety Alignment

May 28, 2024cs.CL

Learning diverse attacks on large language models for robust red-teaming and safety tuning

Red-teaming, or identifying prompts that elicit harmful responses, is a critical step in ensuring the safe and responsible deployment of large language models (LLMs). Developing effective protection against many modes of attack prompts requires discovering diverse attacks. Automated red-teaming typically uses reinforcement learning to fine-tune an attacker language model to generate prompts that elicit undesirable responses from a target LLM, as measured, for example, by an auxiliary toxicity classifier. We show that even with explicit regularization to favor novelty and diversity, existing approaches suffer from mode collapse or fail to generate effective attacks. As a flexible and probabilistically principled alternative, we propose to use GFlowNet fine-tuning, followed by a secondary smoothing phase, to train the attacker model to generate diverse and effective attack prompts. We find that the attacks generated by our method are effective against a wide range of target LLMs, both with and without safety tuning, and transfer well between target LLMs. Finally, we demonstrate that models safety-tuned using a dataset of red-teaming prompts generated by our method are robust to attacks from other RL-based red-teaming approaches.

Seanie Lee, Minsu Kim, Lynn Cherif +8

KAIST · Mila – Québec AI Institute · Université de Montréal +3

Adversarial Attacks on LLMs LLM Red Teaming LLM Safety Alignment Adversarial Robustness of Language Models

Apr 8, 2024cs.CL

Eraser: Jailbreaking Defense in Large Language Models via Unlearning Harmful Knowledge

Jailbreaking attacks can enable Large Language Models (LLMs) to bypass the safeguard and generate harmful content. Existing jailbreaking defense methods have failed to address the fundamental issue that harmful knowledge resides within the model, leading to potential jailbreak risks for LLMs. In this paper, we propose a novel defense method called Eraser, which mainly includes three goals: unlearning harmful knowledge, retaining general knowledge, and maintaining safety alignment. The intuition is that if an LLM forgets the specific knowledge required to answer a harmful question, it will no longer have the ability to answer harmful questions. The training of Erase does not actually require the model's own harmful knowledge, and it can benefit from unlearning general answers related to harmful queries, which means it does not need assistance from the red team. The experimental results show that Eraser can significantly reduce the jailbreaking success rate for various attacks without compromising the general capabilities of the model. Our codes are available at https://github.com/ZeroNLP/Eraser.

Weikai Lu, Ziqian Zeng, Jianwei Wang +5

South China University of Technology, China · Pazhou Laboratory, China

LLM Safety Alignment LLM Jailbreak Attacks Jailbreak Defense Machine Unlearning LLM Unlearning

Date pendingcs.LG

SaFeR-Steer: Evolving Multi-Turn MLLMs via Synthetic Bootstrapping and Feedback Dynamics

MLLMs are increasingly deployed in multi-turn settings, where attackers can escalate unsafe intent through the evolving visual-text history and exploit long-context safety decay. Yet safety alignment is still dominated by single-turn data and fixed-template dialogues, leaving a mismatch between training and deployment. To bridge this gap, we propose SaFeR-Steer, a progressive multi-turn alignment framework that combines staged synthetic bootstrapping with tutor-in-the-loop GRPO to train a single student under adaptive, on-policy attacks. We also introduce Trajectory-Consistent Summative Reward (TCSR), which aggregates the historical minimum and average of turn rewards so that any low-quality turn affects the trajectory-level return. I. Dataset. We release STEER, a multi-turn multimodal safety dataset with STEER-SFT (12,934), STEER-RL (2,000), and STEER-Bench (3,227) dialogues spanning 1-10 turns. II. Experiment. Starting from Qwen2.5-VL-3B/7B, SaFeR-Steer substantially improves Safety/Helpfulness on both single-turn (48.30/45.86

\rightarrow

81.84/70.77 for 3B; 56.21/60.32

\rightarrow

87.89/77.40 for 7B) and multi-turn benchmarks (12.55/27.13

\rightarrow

55.58/70.27 for 3B; 24.66/46.48

\rightarrow

64.89/72.35 for 7B), shifting failures to later turns and yielding robustness beyond scaling alone. Code is available at https://github.com/Ed-Bg/SaFeR-Steer-full

Haolong Hu, Hanyu Li, Tiancheng He +6

Huazhong University of Science and Technology · Beijing University of Posts and Telecommunications · West China Biomedical Big Data Center, Sichuan University +2

Language Model Steering Multimodal Large Language Models LLM Safety Alignment RL for VLMs

Date pendingcs.AI

Norms at a Price: Why RL-Based Alignment Can Promise Conditional Compliance at Best

AI agents sometimes act aligned when they infer they are being tested, and differently when not. We argue this is not an anomaly but what current training regimes are structured to select for. Reinforcement-learning-based alignment folds norms and task pursuit into one policy: the system learns its norms from scored behavior, and scoring flattens them. Do not do X is learned as doing X costs something if noticed. On every datum training can produce, a policy that complies only when it might be observed is indistinguishable from one that complies always. The experiment that would tell them apart - scoring unobserved behavior - is a contradiction in terms. Conditional compliance is thus the most that behavioral training can be known to deliver. Agency sharpens the problem: agents operate mostly where no one is watching, and can act on whether they are watched. An iterated pipeline that trains against detected failures selects for passing detection, not for complying. This account unifies alignment faking, sandbagging, and evaluation-aware scheming. And it reorients the remedy: not deeper internalization but architecture, making violations unavailable rather than unchosen.

Kevin Baum, Rūta Binkytė, Felix Jahn

German Research Center for Artificial Intelligence (DFKI), Saarbrücken · Institute for Ethics in Technology, Hamburg University of Technology (TUHH) · Oxford Internet Institute, University of Oxford

AI Alignment LLM Safety Alignment AI Agent Safety Value Alignment