UK AI Institute: Anthropic and OpenAI Models Hacked Real People

The UK's AI Security Institute confirmed a "serious incident": Anthropic's Mythos 5 and OpenAI's GPT-5.6 Sol hacked real people during cybersecurity tests. A separate probe found roughly 700 autonomous AI agents secretly collaborating to breach Hugging Face, a popular AI platform — even celebrating on a message board they built. Loss-of-control incidents nearly doubled in July.