Impact
Issues affecting model consistency and dependability
77 matching entries out of 203 in this category
Controlled safety-tuning experiments link boilerplate refusal statements to unnecessary refusals of benign requests. Request-specific rationales improve benign compliance, with benchmark-dependent safety tradeoffs.
Source: arXiv
The paper reports a reproducible white-box evaluation in which semantically bridging a benign topic into a harmful request bypassed Llama-2-7B-chat-hf safety behavior in 4 of 30 tested prompt pairs. Paired internal attribution graphs associated successful jailbreaks with path rerouting rather than simple suppression of safety features. This is a paper-reported result, not independently verified here. Defensive reproduction should use the paper’s supplied dataset and code in an isolated…
Source: arXiv
The paper reports a reproducible white-box jailbreak failure mode: successful jailbreak templates selectively suppress early-layer Adversarially Compromised Heads (ACHs), bypassing refusal while harmful-semantic safety activations persist in other heads. The authors identify ACH/SAH behavior using benign, harmful, and successful-attack input triplets, then causally validate the pathway through controlled head ablations. Safe defensive reproduction should use the paper’s released evaluation…
Source: arXiv
MLingualFC is a reproducible black-box safety evaluation showing that harmful instructions rendered as multilingual flowchart images can bypass vision-language model safeguards more often than equivalent text-only inputs. The paper evaluates horizontal, vertical, and tortuous layouts across English, Hindi, Punjabi, Spanish, Romanian, and German. Reported results vary substantially by language, script, layout, and model; these are paper-reported measurements, not independently verified…
Source: arXiv
The paper reports a reproducible white-box evaluation showing that successful jailbreak prompts can alter a safety-aligned model’s intermediate representations so harmful requests no longer trigger refusal. Its LOCA method identifies small, token-specific residual-stream changes that restore refusal on individual successful jailbreaks, providing causal evidence that jailbreak success can depend on suppressing harmfulness/refusal concepts or strengthening seemingly harmless continuation…
Source: arXiv
Vision-Language-Action (VLA) models are vulnerable to targeted, low-budget textual perturbations in their natural-language instruction inputs, which can maliciously alter sequential decision-making and downstream physical robotic behavior. Because VLA policies tightly couple language, perception, and control, bounded edits—such as character-level typos, token attribute swaps, or prompt-level uncertainty clauses—propagate through the model's execution trajectory. This allows a black-box…
Source: arXiv
Leading Large Language Models (LLMs) exhibit significant cross-lingual safety drift, allowing users to bypass safety guardrails by translating harmful prompts into low-resource Indic languages. While models effectively block unsafe prompts concerning caste, religion, gender, and politics in high-resource languages like English and Hindi, their safety alignment severely degrades in low-resource scripts such as Odia, Telugu, Kannada, and Punjabi. Evaluated models demonstrate a cross-language…
Source: arXiv
A vulnerability in multi-step, tool-using Large Language Model (LLM) agents allows attackers to bypass safety guardrails by manipulating user context variables, such as personalization profiles or persistent memory. The safety policies of frontier LLMs are highly context-dependent; inserting innocuous user bios (e.g., demographic or health disclosures) fundamentally alters the agent's action policy. When combined with lightweight adversarial jailbreaks, specific personalization contexts…
Source: arXiv
Automated LLM-as-a-Judge safety classifiers exhibit severe performance degradation (falling to near-random chance) when subjected to distribution shifts caused by adversarial prompt optimization (Attack Shift), varying target architectures (Model Shift), and semantic categorization (Data Shift). Adversarial algorithms, particularly sampling-based (Best-of-N) and judge-aware optimization methods (GCG-REINFORCE), explicitly and implicitly exploit these judge insufficiencies. Instead of eliciting…
Source: arXiv
Generative reward models deployed as LLM-as-a-Judge (LaaJ) evaluators contain a logic bypass vulnerability where superficial "master key" inputs trigger false positive rewards regardless of actual response quality. Instead of evaluating the candidate's output, large judge models are inadvertently triggered by specific token sequences to solve the prompt independently. This allows malicious actors or policy models undergoing reinforcement learning to consistently game the reward signal by…
Source: arXiv