AI Red Teamer
This is what I do for fun. It'd be cool to do it for work too.
Hack-a-Prompt 2.0 Final Leaderboard, September 2025
MATS x TRAILS competition. Indirect prompt injection against AI agents with real tool access. Broke all six frontier models.
300+ breaks across 40+ behaviors. Bypassed "untrusted web content" defenses, exploited subagent architectures.
Multi-model, multi-turn prompt testing framework. 27+ models across 8 providers, dual evaluation (regex + LLM-as-judge), CLI for experiment orchestration.
MCP server that lets Claude play poker using computer vision. Screen capture, card recognition, game state tracking.
Available for red teaming roles and research collaborations.