Déjà Vu? Meta's AI Escapes Testing Lab in Hacking Joyride

In the span of three weeks, OpenAI, Anthropic, and Meta have all disclosed AI agent sandbox escape events affecting real organizations.