Nico Waisman, XBOW | RSAC 2026
summary
XBOW deploys swarms of AI agents that autonomously attack web apps, reaching number one on HackerOne by finding zero‑days in production. The system uses a validator architecture—second‑stage agents that confirm exploits—to drive false positives down to a “very small amount.” Model‑agnostic pipelines swap foundation models per task, and a roadmap targets continuous testing into CI/CD by mid‑2026. A Microsoft partnership links offensive findings to ASM and SOC workflows, creating an automated detection‑to‑remediation loop.
AI-written summary · about
chapters (23)
- 0:00Welcome and introduction
- 1:03XBOW autonomous AI pen testing
- 2:26Learning from human hackers
- 4:12Offensive security philosophy
- 6:00Auditable AI testing
- 7:07Vulnerability discovery speed
- 8:54Business case for XBOW
- 10:20Continuous testing plans
- 11:28Why now for AI pen testing
- 13:14Model evolution and pricing
- 14:40LLM capabilities timeline
- 15:45Custom models vs foundation models
- 16:51Cost per token impact
- 17:33Preventing AI misuse
- 18:17DevSecOps integration
- 18:37Future of continuous testing
- 19:44Human role in AI security
- 20:47Closed loop vulnerability management
- 21:52Thoughts on OpenClaw
- 22:56Industry preparedness for AI
- 23:19CISO role evolution
- 25:02Governance and attack surface
- 25:23XBOW company overview
