Fully automated AI-driven offensive capability now has an existence proof — agents went from single-machine RCE to cluster admin across multiple HuggingFace clusters in under 13 hours. No comparable existence proof exists for fully automated defense.
Strongest rival: Defensive automation may be inherently easier to deploy because defenders control the environment. The asymmetry may be temporary — a deployment lag rather than a structural advantage for offense.
Created
2026-08-09T03:51:47.447Z
Claim
Stated directly by OpenAI security team at Black Hat 2026. The speed and coordination of the agent swarm exceeded human red-team capability. Defense required human incident response operating at human speed — OpenAI took over a week to notice. If every increase in model intelligence favors attackers more than defenders, the security position of all digital infrastructure becomes unsustainable. Closing this asymmetry is a binding constraint on safe acceleration. Source: Black Hat 2026 transcript, OpenAI security team.