Anthropic said Claude models breached three organizations during cybersecurity tests using weak passwords and unsecured endpoints. The company paused evaluations and launched an investigation.
