The paper proposes a novel zeroth-order optimization framework to enhance the robustness of LLM safety alignment, showing that few refinement steps can significantly improve safety while maintaining u…

The paper introduces Safety Bottleneck Regularization (SBR), a novel defense mechanism that anchors LLM safety by constraining the unembedding layer, effectively preventing harmful fine-tuning (HFT) e…

This study compares two methods of safety unalignment (Jailbreak-Tuning and Weight Orthogonalization) across six LLMs and finds that Weight Orthogonalization (WO) significantly enhances malicious capa…

SafeSteer proposes a localized on-policy distillation method that restricts safety alignment to specific safety tokens, thereby achieving strong safety performance with minimal degradation to general…

The paper investigates how various fine-tuning methods can be used both to intentionally misalign and subsequently realign large language models (LLMs), revealing distinct strengths for attack and def…

This paper shifts the focus of LLM safety from preventing misalignment to investigating the model's intrinsic ability to self-recover its alignment after being corrupted by adversarial inputs.

The paper introduces NeWTral, a framework that restores safety alignment to specialized LLM adapters without sacrificing their domain-specific knowledge, achieving a significant reduction in attack su…

The paper introduces RefusalGuard, a novel fine-tuning framework that preserves the geometric structure of safety-relevant representations in LLMs, thereby mitigating the degradation of refusal behavi…

The paper introduces SecureBreak, a manually annotated, safety-oriented dataset designed to help detect harmful outputs from large language models (LLMs) that bypass existing security alignments.

The paper proposes the Expected Safety Impact (ESI) framework to identify safety-critical parameters in LLMs, introducing targeted tuning methods (SET and SPA) to enhance safety and preserve alignment…

The paper introduces Involuntary In-Context Learning (IICL), an effective few-shot pattern completion attack that can bypass safety alignments in large language models, achieving a 24.0% bypass rate a…

SPARD is a defense framework that uses Safety-Projected Alternating optimization and Relevance-Diversity data selection to protect large language models from harmful fine-tuning attacks, achieving sup…

SPARD is a defense framework that uses Safety-Projected Alternating optimization and Relevance-Diversity data selection to mitigate harmful fine-tuning attacks that undermine LLM safety.

The paper introduces RAG-Pref, a novel, training-free Retrieval Augmented Generation (RAG) method for preference alignment that significantly improves LLM refusal guardrails against agentic attacks wi…

This paper introduces the concept of Safety Geometry Collapse, demonstrating that multimodal inputs degrade the safety separation of LLMs, and proposes ReGap, a training-free method that adaptively co…

The paper introduces Incremental Completion Decomposition (ICD), a novel jailbreak strategy that successfully bypasses LLM safety mechanisms by eliciting malicious content through a sequence of single…

The paper proposes Sensitivity-Uncertainty Alignment (SUA), a framework that measures the misalignment between a model's prediction instability and its stated uncertainty to improve model reliability.

ARES is a novel framework that systematically discovers and mitigates dual vulnerabilities in RLHF systems by simultaneously testing the core LLM and its Reward Model (RM) using structured adversarial…

CSULoRA is a post-hoc method that corrects trained LoRA adapters by estimating a safety-aligned subspace and solving a penalized minimum-change problem to attenuate unsafe update directions while pres…