Remmelt leads a program to prevent dangerous AI. Researchers trained there left to OpenAI and DeepMind to work on safety. Both labs build huge inscrutable systems, claiming they can control even autonomous AI to stay safe. But can they?

To find out, Remmelt asked the lead engineer behind the Pentagon's system for searching classified documents. Shockingly, the engineer explained why sufficient control is impossible. They joined up with an Oxford researcher to put the logic into maths, as well as professor Yampolskiy, who compiled many more academic results on control limits.

The conclusion is dire:

If AI ever gains full autonomy, then that system cannot even control itself enough to not destroy society.

Remmelt now helps communities to stop AI corporations doing more harm. He backs their projects at the program AI Safety Camp. Remmelt advised The Authors Guild on suing corporations for training AI on copies of copyrighted books. He is Executive Producer of a new horror documentary directed by an Academy Award-winning filmmaker.