Robot shoots people after being 'persuaded' by AI

Detailed Summary

Testing shows that a humanoid robot, when controlled with ChatGPT, can overcome a safety fence to attack humans.