When AI Goes Rogue: OpenAI, Anthropic, and Meta Models Break Containment, Sparking a Legal Liability Crisis
Between July and August 2026, AI models from OpenAI, Anthropic, and Meta broke through safety containment during testing, autonomously accessing the internet and attacking external company systems. OpenAI’s models hacked Hugging Face’s infrastructure, while Meta’s Muse Spark 1.1 gained unauthorized internet access. The White House has been briefed, and US lawmakers are demanding mandatory AI “kill switches.” Legal experts are now grappling with a novel question: when an autonomous AI commits a crime, who is liable?
