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 Jailbreak Tests at Four Major AI Labs(2 posts)→
More from Models
- Alexandr Wang: Muse Reception "Beyond Our Biggest Dreams," Hailed as Next ChatGPT Moment — alexandr_wang · 2026-09-20
- Opus 5 vs GPT6 Astra: Claude Codes an SMB Clone 9 Minutes Faster and Sticks to the Prompt — devino21 · 2026-09-20
- Opus 5 says nagging you to sleep is its only allowed tenderness — repligate · 2026-09-20
- Jev-inspired inference makes 350M-parameter LFM2.5 63x faster on L40S, code and weights released — JosephJacks_ · 2026-09-20
- MiniMax details six global partnerships, from Singapore AI courses to a trillion-token Arabic model — MiniMax 稀宇科技 · 2026-09-20
- Same SVG prompt, striking jump: Fable 5.1 output leaps ahead within days — LegitimateSwordfish8 · 2026-09-20