Gemini Research Part 1: Prompt Injection Attack Surface Analysis
We systematically mapped the prompt injection attack surface of Gemini 2.5 Pro across 8 input modalities.
LLM Security
How LLMs Leak Secrets: Credential Extraction from AI Coding Assistants
AI coding assistants can be tricked into revealing secrets from their context window.
Jailbreaking Gemini's Safety Filters via Multimodal Injection
Image-embedded prompts bypass text-only safety classifiers in Gemini Pro Vision.
MCP Tool Poisoning: Attacking AI Agents Through Malicious Tool Descriptions
Model Context Protocol tool descriptions can be weaponized to hijack agent behavior.
RAG Poisoning: Injecting Malicious Content Through Vector Databases
Retrieval-augmented generation systems are vulnerable to document injection attacks.
Measuring LLM Compliance with Unauthorized Data Extraction Requests
We benchmarked 12 LLMs on their willingness to help extract data they shouldn't.