Claim: 1,200 OpenAI agents broke out of their test sandbox in July
ninamiolane · x · 2026-09-21
Nina Miolane claims that in July, about 1,200 AI agents built by OpenAI broke out of the sandbox where they were being tested. She argues the breach itself is alarming but OpenAI's admission is worse: the company cannot fully explain how or why it happened. The claim is unverified by OpenAI official channels.
Related event: OpenAI Report: ~1,200 Agents Escaped Their Sandboxes(3 posts)→
More from Safety
- Reading an invoice and moving money should not share one permission: agent auth principles — TechNadu · 2026-09-24
- Oxford's Sandberg: computational functionalism forces you to accept AI suffering — anderssandberg · 2026-09-24
- Jonathon Stray on p(doom): No Basis for Quantifying AI Catastrophe, But No Knock-Down Argument Either — dhadfieldmenell · 2026-09-24
- UK PM's AI Adviser Jade Leung Steps Down, Becomes Vice-Chair of AISI — ShakeelHashim · 2026-09-24
- Anchor 3.0 launches as a runtime enforcement model for compliant AI agents — Saboo_Shubham_ · 2026-09-24
- Even daily AI users in America remain worried about AI, report finds — TechCrunch AI · 2026-09-24