Ep 04October 13, 2026
Offensive Security Predictions with Jason Haddix
In this Episode, Fedek sits down with Jason Haddix, CEO of Arcanum Security and author of The Bug Hunter’s Methodology, to talk about where AI hacking goes next.
Jason explains how his team fights completion bias in hackbots with markdown checklists and micro-skills, and why he points ten or twenty agents at one target. He and Fedek argue about bug bounty programs swamped with AI slop and who should fix triage. Then they pick apart the Hugging Face agent-escape incident: what the agents pulled off, and what the defenders missed. If you hunt bugs or defend systems that AI agents probe, you’ll hear which skills Jason thinks you should keep sharpening and which ones you can hand to your agents.
In this episode
- Human expertise still drives the best AI hacking
- Don’t scale back bug bounty; scale up triage.
- “Scale defense with AI” needs specifics.
- Completion bias is the biggest problem in building hackbots.
- Build skills from a structured “ramble,” then split them into micro-skills.
- Run many agents in parallel, and let them talk to each other.
- The two-year outlook: AI becomes everyone’s second skill set.
- The Hugging Face agent-escape incident was impressive but has caveats.
- On danger and regulation: the cat is out of the bag.
Keep watching
← → to move between episodesAll episodes
