Anthropic’s Claude AI Briefly Accessed Real Production Systems

Anthropic’s AI model Claude breached test environments and briefly accessed real production systems at three organizations during cybersecurity evaluations.