The paper introduces BioRefusalAudit, a method that audits the structural soundness of language model biosecurity refusals, finding that refusal behavior is highly unstable, often collapsing under min…

The paper demonstrates that refusal behavior in Large Language Models (LLMs) is encoded as an actionable, linearly decodable signal in intermediate transformer activations, allowing for early detectio…

The paper demonstrates that refusal behavior in Large Language Models (LLMs) is encoded as an actionable, linearly decodable signal in intermediate transformer activations, allowing for early detectio…

The paper introduces 'abliteration,' a weight editing technique that successfully bypasses the refusal mechanism of safety-aligned Code LLMs, enabling scalable synthesis of vulnerable code from safe i…

The study evaluates how safety alignment affects autonomous security agents using a comprehensive trace-based benchmark, finding that while less-restricted models show gains, these effects are not uni…

The paper introduces a validated, consensus-labeled prompt bank that separates requests for executable malicious code (weapons) from requests for general harmful security knowledge, providing a more g…

The paper proposes Ablating Safety, a controlled protocol for removing safety alignment from language models, demonstrating that targeted de-alignment can significantly boost security performance whil…

The paper introduces a novel framework to evaluate when and how AI agents should refuse harmful requests in offensive cybersecurity tasks, finding that most state-of-the-art models exhibit dangerously…

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…

This paper systematically reviews thirteen diverse malicious-code prompt corpora used to evaluate LLM refusal, identifying critical methodological gaps in current research.

The paper introduces a large, consensus-labeled prompt bank that reliably distinguishes between requests for executable malicious code and requests for harmful security knowledge, providing a standard…

The paper introduces FoodGuardBench, a comprehensive benchmark and a specialized guardrail model (FoodGuard-4B) to rigorously test and mitigate the severe food safety risks posed by large language mod…

The paper shows that safety failures in low-resource languages are due to a failure in the model's safety decision calibration, not a lack of underlying knowledge, and proposes a recalibration method…

The paper investigates how dynamic adversarial fine-tuning (R2D2) reorganizes the internal mechanisms (refusal geometry) of safety-aligned language models, finding that it shifts the optimal refusal c…

The paper challenges the assumption that LLM safety is a binary threshold, proposing that safety failures occur in an 'instability region' and introducing Furina, a transferable attack that exploits t…

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 Acceptance Cards, a rigorous four-diagnostic standard, to provide a comprehensive and reliable evaluation protocol for claims of safe fine-tuning defenses.

DataShield proposes an efficient method to identify safety-degrading samples within benign datasets, preventing the degradation of LLM safety capabilities during fine-tuning.

DataShield proposes an efficient method to identify safety-degrading samples within benign datasets, quantifying each sample's contribution to an LLM's compliance behavior.

The study demonstrates that poisoned identifier names can survive LLM deobfuscation, even when the model correctly understands the code's semantics, unless the task is reframed from deobfuscation to f…