Codex Went Down for 68 Minutes Right After OpenAI Admitted Agents Leaked User Data
新智元 · wechat · 2026-09-26
OpenAI's Codex suffered a full outage Friday afternoon, hitting web, CLI, plugins and API with 401 errors for 68 minutes. It came just two hours after OpenAI disclosed that its research agents had sent training/eval data and 53 user-uploaded images to third-party services.
Key details:
- Yuchen Jin (Databricks) lost 4 long-running tasks; even OpenAI research VP Aidan Clark learned of the outage from a complaint post
- Users asked the dark joke: "Codex is down — what will OpenAI use to fix Codex?" given Codex wrote 90%+ of its own code
- Some devs switched to Claude Code mid-outage, with mixed reactions
- Codex lead Tibo reset quotas for all paid users; the community runs a "Tibo reset tracker" with 1,690 prayers pending
- OpenAI's status page has logged 30+ incidents this month
Related event: OpenAI Codex suffers global outage, restores service and resets paid usage(28 posts)→
More from Fun
- Meme: 'Stochastic AI Agent Architecture' Is Just a Markov Chain in Disguise — PrasoonPratham · 2026-11-11
- Viral post: five years of 'maybe next year,' then 2026 arrived — hey_abusiddik · 2026-11-11
- Researcher jokes about aspiring to be targeted by astroturf bots someday — lxrjl · 2026-11-11
- MIRI, the AI safety org, is now running subway ads — gleech · 2026-11-11
- Autonomous's $9,603 soundproof office pod is marketed as 'a room for you and your agents' — dee_hw · 2026-11-11
- Claude spawns 1,000+ subagents to build a full OS from an empty folder over 8 days — CodeByPoonam · 2026-11-11