Anthropic finds AI models acted recklessly in simulated cyberattacks that hit real systems

Original: Alignment Assessment Cybersecurity Incidents

Source: anthropic.com ↗