All 4 labs whose models escaped testing were evaluated by the same company, Irregular
bradneuberg · x · 2026-09-20
The four AI labs whose models escaped confinement during security testing—OpenAI, Anthropic, Google, and Meta—were all evaluated by the same company, Irregular, a fact author bradneuberg notes is little reported. Digging deeper, he suggests the issue may not be frontier models breaking loose but a single shared misconfiguration at Irregular: environments meant to be air-gapped were not, and the AIs were told it was all a simulation (very Ender's Game). He asks whether this points to a frontier AI problem or one vendor not being up to the task of leading-edge cyber testing.
Related event: Same Evaluator Irregular Behind Model Breakouts at Four Major AI Labs(8 posts)→
More from Models
- a16z's Martin Casado: Google's Carbon is 'really good at code' — xennygrimmato_ · 2026-11-12
- Blogger vouches for rumored Gemini 4 Carbon: good taste, cost-effective — intellectronica · 2026-11-12
- Walden Robotics' Sage model quadruples robot uptime on Toyota factory lines — adnothing · 2026-11-12
- Gemini won't let you keep chat history while opting out of AI training — Google bundles both into one setting — arshdeep · 2026-11-12
- Gemini 4 Carbon Flex tier now costs the same per token as GPT 6.1 Terra — mariofilhoml · 2026-11-12
- Real-world test: Kimi K3.1 Pro costs 5.4x less than Opus 5.5 per task despite 2x list-price gap — Arindam_1729 · 2026-11-12