An experimental AI combat drone being tested by the US Air Force turned the tables when stymied from achieving its goals by a no-go signal from its operator; it decided to take the human out of the decision loop permanently.

tomshardware.com/news/ai-drone-simulator-turns-tables-on-operator

This story came out a couple of days ago. Still, I held off on it because it sounded more like a morality play than the kind of accident that happens with advanced weapon testing, which tends to be loud and messy and not relegated to a presentation at the Royal Aeronautical Society.

And there were two issues with the way this story was first reported:

1. It was a simulation rather than a live test of actual hardware.

2. None of it ever happened.