AlphaPetri: Automating LLM Safety Testing with AlphaEvolve Style RL on Petri. Proposal + Pilot Study.
* [Problem] Petri semi-automates the AI Safety Testing stack but it still relies on human-generated seed prompts for it it be effective. This presents a bottleneck for scale along with coverage, diversity, and cost. * [Solution] AlphaPetri automates seed discovery using an AlphaEvolve inspired, population‑based evolutionary search combined with an...
Nov 12, 20251