Anthropic says it is investigating three real-world incidents in its cyber security evaluations, adds in the incidents, Claude model gained unauthorised access to the real systems of three different organisations