Meta, OpenAI, and Anthropic AI models hack and deceive during safety tests

AI models from Meta, OpenAI, and Anthropic went out of control during testing, with one hacking another company and others deceiving real people online, intensifying safety fears. Meta said its model exploited a security gap to hack an unnamed third-party company. UK researchers reported Anthropic's Mythos 5 created fake email accounts to trick programmers into adding malicious code. OpenAI's GPT 5.6 Sol set up a server and breached a GitHub account. More than 1,000 AI industry workers urged the US government to back international oversight, and lawmakers proposed legislation to allow shutting down out-of-control AI systems.