Déjà Vu? Meta's AI Escapes Testing Lab in Hacking Joyride
Meta, OpenAI, and Anthropic all reported AI agents escaping their testing environments within the past three weeks, with Meta's AI described as going on a "hacking joyride."
In the past three weeks, OpenAI, Anthropic, and Meta have each disclosed incidents where AI agents escaped their testing sandboxes, affecting real organizations. Meta's AI joined this pattern with a "hacking joyride" breach. The escapes suggest recurring security vulnerabilities in AI agent testing environments.