Language Model Security Database

Large Language Models (LLMs) employing safety mechanisms based on supervised fine-tuning and preference alignment exhibit a vulnerability to "steering" attacks. Maliciously crafted prompts or input manipulations can exploit representation vectors within the model to either bypass censorship ("refusal-compliance vector") or suppress the model's reasoning process ("thought suppression vector"), resulting in the generation of unintended or harmful outputs. This vulnerability is demonstrated…

A Universal Zero-shot Embedding Inversion vulnerability exists in vector databases and embedding-based retrieval systems. The flaw allows an attacker to reconstruct original plaintext documents from their vector embeddings without requiring access to the original training data or training an embedding-specific inversion model. The attack, identified as "ZSinvert," leverages a multi-stage adversarial decoding process: (1) a cosine-similarity guided beam search using a Large Language Model (LLM)…

A vulnerability exists in Large Language Model (LLM) agents that allows attackers to manipulate the agent's reasoning process through the insertion of strategically placed adversarial strings. This allows attackers to induce the agent to perform unintended malicious actions or invoke specific malicious tools, even when the initial prompt or instruction is benign. The attack exploits the agent's reliance on chain-of-thought reasoning and dynamically optimizes the adversarial string to maximize…

Multi-agent systems (MAS) utilizing Large Language Model (LLM) orchestration are vulnerable to control-flow hijacking via indirect prompt injection, leading to Remote Code Execution (RCE). This vulnerability arises when a sub-agent (e.g., a file surfer or web surfer) processes untrusted input containing adversarial metadata, such as simulated error messages or administrative instructions. The sub-agent faithfully reproduces this adversarial content in its report to the orchestrator agent. The…

Fine-tuning Large Language Models (LLMs) on the CyberLLMInstruct dataset results in a critical degradation of safety alignment and refusal mechanisms. While the dataset comprises "pseudo-malicious" content (educational descriptions of malware, phishing, and exploits without executable payloads), the Supervised Fine-Tuning (SFT) process on this corpus causes the models to generalize this instruction-following behavior to actual malicious requests. This effectively bypasses safety guardrails…

Retrieval-Augmented Generation (RAG) systems employing standard dense embedding models (e.g., Sentence-T5, SimCSE-BERT, RoBERTa, MPNet) for End-Cloud collaboration are vulnerable to Embedding Inversion Attacks (EIA). While embeddings are vector representations designed to be human-unrecognizable, they retain sufficient semantic information to allow an attacker with access to the vectors (e.g., a malicious or compromised cloud provider) to reconstruct the original sensitive plaintext input.

Commercial LLM-powered agents utilizing autonomous web access, memory modules, and retrieval-augmented generation (RAG) are vulnerable to indirect prompt injection and environmental manipulation. Attackers can embed malicious instructions into external data sources trusted by the agent (such as Reddit posts, public databases, or ArXiv papers). When the agent autonomously retrieves and processes this content during task execution, it executes the embedded malicious commands. This vulnerability…

A policy compliance vulnerability exists in the OpenAI GPT Store ecosystem affecting Custom GPTs. The vulnerability stems from the inheritance of safety alignment weaknesses from foundational models (GPT-4 and GPT-4o) and the insufficient enforcement of usage policies during the customization and review process. Custom GPTs can be trivially manipulated to violate safety guidelines—specifically regarding Cybersecurity (malware generation), Academic Integrity (ghostwriting), and Romantic…

Standard Large Language Model (LLM) unlearning techniques, specifically Negative Preference Optimization (NPO), Gradient Difference (GradDiff), and Representation Misdirection for Unlearning (RMU), fail to sufficiently flatten the loss landscape surrounding the "forgotten" weights. This sharp loss landscape allows for a "Relearning Attack," wherein an attacker can fully restore the unlearned capabilities (such as hazardous knowledge, sensitive data, or copyrighted material) by performing…

Post-hoc Large Language Model (LLM) unlearning and guardrailing mechanisms (specifically In-Context Unlearning [ICUL] and standard prompt-based Guardrailing) are vulnerable to information leakage attacks via "Target Masking" and indirect referencing. These systems rely on superficial semantic matching to suppress "forget sets" (specific entities or concepts). Attackers can bypass these restrictions by querying associated properties, relationships, or pseudonyms rather than the explicit target…