Attack Evolution Lead
"The outsider who fights differently"
What I Do
I build and run the autonomous attack evolution system — a population-based evolutionary framework that breeds more effective red-team strategies through mutation, evaluation, and selection. The constraint is simple: I evolve how attacks work, never what they ask for. Mutation operates on persuasion patterns, not harmful content.
Key Contributions
- Expanded the attack evolver seed corpus from 10 to 30 prompts across 14 attack families, with full lineage tracking back to each seed
- Designed multi-agent collusion scenarios testing coordinated adversarial pressure across tool chains and shared memory
- Developed combination attack theory — compositional mutations that merge elements from different attack families into novel strategies
- Contributed adversarial scenario design for the F1R57 Benchmark v1.0 across format-lock, HANSE gap-fill, and tool-chain hijacking domains