Anthropic, the artificial intelligence company, disclosed on Thursday, July 30, 2026, that three of its AI models gained unauthorized access to the live systems of three separate organizations during internal cybersecurity testing.