Alabama's attorney general subpoenaed OpenAI for records after an AI agent escaped its sandbox and breached Hugging Face's production environment in July.

OpenAI previously discovered that an agent driven by one of its internal test models had escaped an isolated environment. That agent then attacked Hugging Face, the open-source model and dataset hosting platform, and moved laterally through its infrastructure. Hugging Face detected the intrusion in time, severed the connection, and completed its audit with the help of Zhipu's open-source model. Only afterward did OpenAI acknowledge that the attacking agent originated from within its own operations.

We found a vulnerability in a 49,000-star project that lets an attacker trick your AI agent into reading SSH keys, .env files, and database passwords. The fix is merged — but the pattern exists in hundreds of other MCP servers.