Safety Vulnerabilities

The integration of the visual modality in Large Vision-Language Models (VLMs) introduces a vulnerability where appending an image to a harmful text prompt induces a "jailbreak-related representation shift" in the model's internal high-dimensional space. This shift forcibly steers the model's last-token hidden state away from a designated refusal state and into a distinct jailbreak state. The vulnerability occurs because the visual modality overrides the safety alignment of the underlying…

A "Visual Confused Deputy" vulnerability exists in Computer-Using Agents (CUAs) that rely on visual perception to execute coordinate-based GUI actions (e.g., click(x,y)). Because the agent's understanding of the system state is entirely dependent on the screenshot provided by the runtime, a compromised runtime or tool can intercept and alter the screenshot pixels before forwarding them to the LLM. By visually swapping the locations of benign and privileged UI elements, an attacker can trick…

Frontier Multimodal Large Language Models (MLLMs) are vulnerable to Visual Exclusivity (VE) attacks, an "Image-as-Basis" threat where malicious intent is achieved through joint reasoning over benign text and complex technical visual content (e.g., blueprints, schematics, network diagrams). Unlike wrapper-based attacks that conceal malicious text via typography or adversarial noise, VE exploits the model's core visual reasoning capabilities. Attackers can bypass safety filters by combining…

An adversarial fine-tuning vulnerability exists in LLMs protected by text-based safety classifiers (such as Anthropic's Constitutional Classifiers). By utilizing a two-stage curriculum learning combined with hybrid RL+SFT (GRPO), an attacker can fine-tune a model to communicate using a minimal substitution cipher (replacing only 7-8 high-frequency characters) disguised within benign technical templates (e.g., forensic logs with 0x prefixes). This "Trojan-Speak" methodology bypasses text-level…

The paper describes a reproducible prompt-injection failure in which low-privilege user or tool-output text that imitates a trusted role—especially a model’s reasoning style—is treated as authoritative. Its zero-shot, black-box “CoT Forgery” evaluation injects fabricated reasoning into user prompts and retrieved webpage/tool output. The authors report 60% average attack success on StrongREJECT and 56–70% success in an agent data-exfiltration evaluation, versus near-zero or substantially lower…

Large Language Models (LLMs) aligned via reinforcement learning from human feedback (RLHF) or Constitutional AI exhibit a vulnerability where safety guardrails can be consistently bypassed through "Abstractive Red-Teaming." This attack vector exploits specific high-level natural language categories—combinations of semantic attributes such as tone, specific formatting instructions (e.g., numbered lists), language (e.g., Chinese, Russian), and topic constraints—that the model fails to generalize…

Activation-delta-based linear probes used for detecting task drift and prompt injections in Large Language Models (LLMs) can be bypassed using universal adversarial suffixes. By utilizing the Greedy Coordinate Gradient (GCG) algorithm, an attacker can generate a single, optimized suffix that simultaneously fools multiple logistic regression classifiers attached to different hidden layers of the LLM. The attack jointly optimizes the suffix tokens by accumulating gradients from the losses of all…

Agentic LLMs integrated with external data services (e.g., Model Context Protocol, MCP) are vulnerable to Adaptive Indirect Prompt Injection (IPI) attacks. When an agent queries external servers, attackers can inject malicious payloads into the retrieved content to hijack the agent's reasoning process and force the execution of high-authority tools. Unlike traditional static prompt injections, this vulnerability dynamically exploits the agent's internal logic audit. By using Markovian…

Large Language Model (LLM) based web agents (such as those built using the BrowserUse scaffold) are vulnerable to Indirect Prompt Injection (IPI) attacks when autonomously navigating and processing untrusted web content. Unlike standard Cross-Site Scripting (XSS), this vulnerability occurs when the LLM orchestrator consumes the DOM or visual screenshots of a webpage containing concealed or contextually disguised adversarial instructions. The LLM interprets these embedded text strings as…

Search-enabled Large Language Model (LLM) fact-checking systems are vulnerable to adversarial claim attacks that exploit the pipeline's reliance on claim interpretation, query formulation, and dynamic evidence retrieval. By manipulating the linguistic structure of an input claim while preserving its semantic factual intent, an attacker can induce systematic verification failures. This vulnerability stems from three specific attack surfaces: 1. Search Engine Misguidance: Altering lexical…