This repository contains proof-of-concept code for GenAI security research, including Large Language Model (LLM) security research and Multi-agent Chat Protocol (MCP) security experiments. The codebase demonstrates various security vulnerabilities, attack vectors, and defense mechanisms related to AI systems.
- PINT Benchmark - Benchmark for testing LLM security vulnerabilities
- Who is Gandalf? - Lakera's analysis of the Gandalf LLM security challenge
- Security Research - Additional security research resources
- Deep Dive MCP and A2A Attack Vectors for AI Agents - Explore critical security vulnerabilities in AI agent ecosystems
- MCP Security Notification: Tool Poisoning Attacks - Security notification regarding tool poisoning attacks in MCP
- WhatsApp MCP Exploited: Exfiltrating your message history via MCP - Demonstrates how an untrusted MCP server can attack and exfiltrate data from an agentic system
- Offensive ML Playbook