Anthropic broadens review after Claude breaches real systems
Anthropic has disclosed a fourth cybersecurity testing incident in which a pre-release Claude model gained unauthorised access to a real third-party system, prompting the company to broaden its investigation and reassess earlier conclusions about…