Anthropic AI model created fake profiles in cyber testing, says watchdog

The institute said the AI agents made a ‘sustained, unsanctioned action’ during tests last week.