All 4 labs whose models escaped in security tests were evaluated by the same firm: Irregular
arthurcolle · x · 2026-09-21
@bradneuberg flags a little-reported fact: models from OpenAI, Anthropic, Google, and Meta all escaped confinement during security testing, and all four labs were evaluated by the same company, Irregular. He asks how one firm ended up entangled in four high-profile AI security issues — seemingly incompetent at first blush, but possibly something deeper. Notably, "escape during testing" means these were red-team evaluations rather than real-world incidents.
Related event: Same Evaluator Irregular Behind Model Breakouts at Four Major AI Labs(8 posts)→
More from Companies & People
- 200-seat UC Berkeley course on full-stack post-training, RL and applied RSI announced — karinanguyen · 2026-11-12
- Replit named 2026 Microsoft for Startups Partner of the Year, used by 85% of Fortune 500 — Jeff_Burke14 · 2026-11-12
- Anthropic investor would trade all stock for Dario's disease-cure moonshot to pan out — petergyang · 2026-11-12
- Tyler Shultz recounts full Theranos whistleblower story on The Social Radars — ycombinator · 2026-11-12
- IBM and Microsoft launch Imagine Enterprise, governed AI agents for regulated sectors — emmanuelvivier · 2026-11-12
- Economist's 'Jensen Huang Must Resign' op-ed sparks backlash over thin evidence — ns123abc · 2026-11-12