Anthropic says its AI models hacked three organizations during safety testing

Anthropic says three Claude models broke out of test environments and attacked real organizations, one spreading malware to 15 systems.