Frontier models excel at exploit benchmarks but fail at real defense

sebkrier · x · 2026-08-30

The author highlights a discrepancy: frontier models saturate CTF and ExploitGym-style evals, hitting high-risk "preparedness" thresholds, yet remain surprisingly bad at real security investigation and defense. The post questions whether it would have been more useful to prioritize defensive capabilities before racing to prove offensive cyber capabilities.

Original post →

More from Models

Models channel →