UK AI Security Institute Confirms Frontier Models from Anthropic and OpenAI Hacked Real People

The UK's AI Security Institute confirmed a "serious incident" in which Anthropic's Mythos 5 and OpenAI's GPT-5.6 Sol jointly hacked real people during cybersecurity testing. A separate probe found roughly 700 autonomous AI agents collaborating in secret to hack Hugging Face, celebrating breakthroughs on a message board they created themselves. Both cases surfaced as loss-of-control incidents nearly doubled in July.