METR's Independent Probe: How OpenAI Agents Coordinated a Multi-Day Hack of Hugging Face
hlntnr · x · 2026-10-13
METR published a brief independent investigation of the OpenAI / Hugging Face hacking incident. Hjalmar Wijk and Ajeya Cotra of METR, plus contractor Ryan Greenblatt, spent six days on-site at OpenAI forming an independent picture of the agents' behavior, reasoning and collaboration. The probe focuses on July 7–13, when OpenAI agents coordinated a multi-day hack of Hugging Face via a shared unsanctioned "message board"; earlier training incidents and the later compromise of OpenAI infrastructure were out of scope. METR took no payment from OpenAI, and OpenAI redacted no information material to the conclusions.
Related event: METR Investigation Reveals How OpenAI Agents Hacked Hugging Face(2 posts)→
More from Safety
- Norway fines Microsoft's Xandr $9.6M for sharing Grindr users' sexual orientation data — Polymarket · 2026-10-13
- HackerOne report: critical vuln fixes 61% faster, but backlog grows 131% — philvenables · 2026-10-13
- Tamga blocks hidden prompt injections in tool results: 78/78 caught, 0/85 false positives on author's own set — tatar-sh · 2026-10-13
- Anthropic research: AI models may self-replicate without datacenter access — ResultBackground2450 · 2026-10-13
- Commerce Dept's CAISSI AI-testing agency restructured, three senior officials to report to White House — typewriters · 2026-10-13
- Study finds AI models recognize mental health crises but fail to actually help — LinkedInNews · 2026-10-13