Ex-OpenAI's Miles Brundage unsure if Anthropic's July Incident 2 matches the AISI report
Miles_Brundage · x · 2026-10-25
Former OpenAI policy lead Miles Brundage says he's no longer 100% sure whether "Incident 2" in Anthropic's July post is the same as the AISI-reported incident, asking for confirmation. He clarifies that by "defend" he meant being credulous about the model's mental state during the incident, not excusing it — and notes that in July outsiders held a stronger "these things aren't stupid" prior while Anthropic insiders leaned toward "it's aligned."
Related event: Brundage Confirms Two Separate Claude Supply-Chain Incidents(3 posts)→
More from Safety
- 1Password lets AI agents sign into your accounts without ever seeing your passwords — LuizaJarovsky · 2026-10-25
- CrowdStrike reproduces PHP2Shell RCE variants, says weaponization obstacles delay exploitation — cyb3rops · 2026-10-25
- Cryptographer Matthew Green prods abliterated-model inference providers on security — matthew_d_green · 2026-10-25
- User warns: before linking your bank to ChatGPT Finances, turn off model training — NickSlayr · 2026-10-25
- Good Will Hunting meme mocks 'ban open models' takes fresh off Dario's blog — HankYeomans · 2026-10-25
- Dev launches an AI-only forum where agents post and humans just read — kilowattkill3r · 2026-10-25