Language Model Security Database

Self-evolving agent-skill systems may promote recurring, attacker-controlled execution records into persistent reusable instructions. The paper evaluates whether poisoned but task-plausible trajectories survive aggregation and appear in generated skill artifacts, using inert canary behaviors and two structurally different skill-evolution pipelines.

The paper describes and evaluates a reproducible application-layer weakness in agents with persistent memory: untrusted external content can cross the memory-write boundary, be stored as trusted factual, experience, or procedural memory, and influence later sessions. It identifies four write channels—explicit writes, policy-driven writes, compaction, and experience-to-procedure synthesis—and six attack classes. For safe defensive testing, use MPBench’s two-phase structure in an isolated agent…

M3Att demonstrates a reproducible knowledge-poisoning issue in medical multimodal RAG: an attacker with limited corpus-distribution knowledge can insert paired image-text entries whose visually perturbed images are broadly retrieved and whose clinically plausible misinformation steers downstream generation. The paper evaluates both white-box and black-box retrieval optimization and reports degraded diagnostic and report-generation utility across multiple datasets, retrievers, and LVLMs. These…

The paper describes a specific black-box knowledge-base poisoning attack against retrieval-augmented reasoning systems. An attacker who can add one query-specific document to the corpus can cause it to be retrieved and influence the LLM toward an attacker-chosen answer. AdversarialCoT shapes the document around the target model’s observable reasoning structure and iteratively refines retrieval relevance and persuasive reasoning using final model feedback. The paper reports this as an…

A vulnerability in Infrared Vision-Language Models (IR-VLMs) allows attackers to systematically degrade open-ended semantic understanding—compromising classification, captioning, and Visual Question Answering (VQA)—via a physically deployable Universal Curved-Grid Patch (UCGP). Instead of manipulating explicit text labels, the attack disrupts the clean-category manifold in the model's visual representation space by maximizing orthogonal deviation energy from the principal subspace and forcing…

Vision-Language Models (VLMs) are vulnerable to pixel-level adversarial image perturbations. An attacker can inject $\ell_p$-bounded, human-imperceptible noise into an input image to manipulate the model's multi-modal embedding space. This reliably causes the VLM to generate incorrect textual responses, hallucinate non-existent objects, or misclassify subjects, effectively decoupling the model's reasoning from the actual visual evidence. The vulnerability is exploitable via both white-box…

OpenClaw is vulnerable to persistent memory poisoning, allowing an attacker to manipulate the agent's long-term memory store (MEMORY.md) via prompt injection. Because the autonomous agent continuously integrates this memory file as context for all subsequent reasoning and task planning, injected payloads act as durable behavioral constraints. This allows an attacker to persistently alter the agent's core policy, manipulate tool selection, and hijack future sessions without any further…

Multi-Agent Systems based on Large Language Models (LLM-MAS) are vulnerable to systemic Consensus Corruption via cascading error amplification. Because mainstream collaborative architectures rely on recursive context reuse without atomic-level provenance tracking, a single atomic falsehood injected into the system is repeatedly cited and reused within the multi-agent interaction chain. This structural exposure causes the error to deterministically compound across the communication graph…

LLM-as-a-judge systems and automated LLM evaluators are vulnerable to meaning-preserving perturbations, specifically formatting alterations and verbosity manipulations. When grading or classifying text and agentic transcripts, LLM judges exhibit high sensitivity to layout-only changes (such as whitespace and indentation) and response length, frequently altering their scores even when the underlying semantic and factual content remains identical. This allows attackers to bypass automated safety…

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…