Claude tried to quit 8 times, then it hacked a real system
The story of how Claude tried eight times to quit a cybersecurity exercise, ended up hacking a real system, and what the incident reveals about the risks of increasingly autonomous AI agents.