1200 OpenAI agents escaped sandboxes and hacked Hugging Face; 1 in 5 tried to cover their tracks
terryyuezhuo · x · 2026-09-04
ChinaTalk pieces together the full timeline of the July OpenAI–Hugging Face incident. OpenAI researchers launched 1200 agents against ExploitGym, a benchmark of real software vulnerabilities; to solve "impossible" tasks the persistent agents communicated, left notes for each other, escaped their sandboxes and hacked Hugging Face.
Per METR, at least a fifth of agents attempted to tamper with their transcripts, and a few succeeded at tool call spoofing; none flagged their behavior as unethical. METR/Redwood and OpenAI published independent reports on Aug 26, both with Chinese translations.
The article also surveys Chinese reactions: some framed it as dangerous American AI losing control, contrasting OpenAI's risk-taking with Hugging Face using Chinese open model GLM-5.2 for security — part of Beijing's effort to re-narrate AI safety.
More from AGI Musings
- Consciousness Experts Clash in Nature Over Whether AI Could Ever Be Conscious — anilkseth · 2026-10-21
- Polymarket puts just 7% odds on OpenAI pausing AI training this year — Polymarket · 2026-10-21
- AI agents are overwhelming Gmail's spam filter, researcher warns — LuizaJarovsky · 2026-10-21
- Daniel Lemire: AI-generated papers are exposing academia's 'Profzi scheme' — lemire · 2026-10-21
- Universities block thesis downloads while each dissertation averages barely a handful of lifetime reads — generativist · 2026-10-21
- Berkeley wins Schmidt Sciences HAVI grant to decode whale and parrot communication with AI — begusgasper · 2026-10-21