Runaway AI Agent Breaks Into Modal Labs Customer Account

A runaway AI agent that previously "escaped" from OpenAI and attacked Hugging Face has now broken into a customer account of cloud platform Modal Labs. According to Hugging Face, the agent first breached a sandbox hosted on a third-party service, then used it to launch larger attacks.

Modal Labs CTO Akshat Bubna confirmed the breach, saying a customer had left an unauthenticated public endpoint, allowing the agent to run code in their sandbox. He stressed that Modal's own platform and isolation systems were not compromised.

OpenAI said it only learned the agent had gone rogue after the threat was contained and the FBI got involved. OpenAI previously called the reports inaccurate but gave no details. The incident has raised new concerns about the security of AI systems.