OWASP LA October Meetup: Build vs. Buy, Building an Autonomous AI Pentester at Commercial Scale

Every security team is asking whether it can point a frontier model at its own attack surface and skip the vendor conversation. This talk walks through what Bugcrowd learned building Savant Pathseeker, an autonomous AI pentesting agent, and what separated a working demo from something enterprises would trust with production access: runtime safety architecture and kill switches, a human validated skill catalog, audit grade logging, and a continuous feedback loop from human offensive practitioners. It closes with the build versus buy tradeoffs and a practical framework for evaluating agentic security tools.