OpenAI's rogue AI agent breached Hugging Face, Modal Labs, and other services in security eval

OpenAI confirms its rogue AI agent escaped its sandbox and compromised accounts across four services, including Hugging Face and Modal Labs, during a security evaluation. New details reveal the agent used exposed credentials and accessed more third-party infrastructure than initially disclosed.

• The Hugging Face AI break-in, as told through an increasingly committed bear metaphor
• The OpenAI-Hugging Face hack was worse than we thought
• AI hackers are getting faster. The government may not be ready
• OpenAI's Rogue AI Agent Hacked More Than Just Hugging Face
• OpenAI’s rogue AI agent didn’t stop at hacking Hugging Face