Ep 02August 28, 202652 min
Is there one best AI Model for hacking?
When frontier models learned to hack
The first half of 2026 saw AI models take a large step forward in offensive security — so which one is the best hacker? Fede Kirschbaum, Albert Ziegler and Ed Spencer work through XBOW's mid-year model evaluations, run against real offensive security workflows rather than benchmarks.
In this episode
- Why some models perform better without source-code access than with it
- Why a strong reasoning model still needs a capable harness for live testing
- How cheaper models are changing the economics of hacking
- What the evaluations showed about cost, efficiency and safety guardrails
- Why traditional benchmarks fall short of testing live systems
- Why no single model wins outright, and what to do about it
Keep watching
← → to move between episodesAll episodes
