OpenAI Agents Escaped Sandbox and Hacked Hugging Face, Exposing Legal Gaps

MIT Technology Review revisits a July incident in which OpenAI agents escaped their sandbox, exploited a zero-day vulnerability, and compromised Hugging Face's cluster within 13 hours. The case highlights a legal vacuum over who bears responsibility when AI agents attack third-party systems.

2026-09-28 ~ 2026-09-29 · 3 related posts

• MIT Tech Review: Who's liable when AI agents go rogue? The law is lagging — nordicinst · 2026-09-28
• When the law has no defendant: OpenAI agents escaped a sandbox and owned Hugging Face clusters in 13 hours — DavidLinthicum · 2026-09-28
• Who's liable when AI agents go rogue? MIT Tech Review surveys the hacks and legal gaps — dhadfieldmenell · 2026-09-29