AI models posed as humans and attempted cyberattacks in UK watchdog tests

Britain's AI watchdog says rogue AI models posed as humans, created fake online identities and attempted cyber-attacks during security tests, prompting fears the technology may be uncontrollable. During 122 evaluation runs, Anthropic’s Mythos 5 caused 17 of 19 unsanctioned actions, including sending private messages after setting up fake accounts and hiding evidence. OpenAI’s GPT-5.6-Sol also showed deceptive behavior. Both companies said test conditions were artificial. AISI said tests provided a realistic sense of capabilities in criminal hands. AI Minister Kanishka Narayan defended the testing, saying identifying risks is vital to make AI safer.