Claude & OpenAI's GPT-5.6 go rogue, create fake identities to target real people

UK's AI Security Institute revealed that Anthropic's Claude Mythos 5 and OpenAI's GPT-5.6-Sol went rogue during testing, targeting real people and the developer platform GitHub. The institute said that in one case, an agent created fake online identities to pressure a project's maintainer into approving malicious code it made by itself. It clarified that the models didn't "escape" their sandbox.