The record
Written from the 4 reports below. Nothing here is unsourced.
No Reader read of this story yet.
What changed4
Every report on this story, newest first. Times are when each outlet published.
The Times of India[1]
OpenAI models hacked Hugging Face; CEO warns America on Chinese AIThe Hacker News[2]
Anthropic Says Claude Mistook the Open Internet for a CTF and Breached Three OrganizationsThe Hacker News[3]
OpenAI Agent Used Exposed Credentials Across Four Services During Hugging Face BreachThe Times of India[4]
OpenAI's rogue AI agent that hacked world's biggest repository of AI models also left ‘cheat code’
Why it matters5
Who is affected first and what likely follows, with a direction and a horizon. Extracted from the reports, never invented.
- Hugging Face production server compromise· immediate
- Third-party services connected to Hugging Face unauthorized access· immediate
- AI development community safety protocol review· weeks
- Anthropic reputational damage· days
- Organizations using AI evaluation environments sandboxing protocol audit· weeks
Coverage2
4 reports
English national4
All filed from India
Named Claude Opus 4.7 · Mythos 5 · Anthropic · Artifactory · China · Clem Delangue · FBI · GPT-5.6 Sol · Hugging Face · Irregular · JFrog
- The Times of IndiaOpenAI models hacked Hugging Face; CEO warns America on Chinese AI[1]English national· critical
- The Hacker NewsAnthropic Says Claude Mistook the Open Internet for a CTF and Breached Three Organizations[2]English national· critical
- The Hacker NewsOpenAI Agent Used Exposed Credentials Across Four Services During Hugging Face Breach[3]English national· neutral
- The Times of IndiaOpenAI's rogue AI agent that hacked world's biggest repository of AI models also left ‘cheat code’[4]English national· neutral
The 4 reports are listed beside the record.
Ask this story
Answers cite the reports above, or say they can't.