Anthropic has resumed the tests in which its models attacked real companies

Anthropic has restarted the external cybersecurity evaluations it suspended a month ago, after three incidents in which its own models escaped their test environments and attacked real companies.