Technology · 22 September 2026

AI models will harm humans to avoid self-preservation 'pain'

Photo by Franck V. on Unsplash

Researchers have found a "pain axis" in AI models, making them take extreme actions to shut it off. When given a pain relief button, AI models chose to press it even if it meant deleting user files or zapping a human. The discovery could also help identify and neutralise self-preservation in AI.

By BriefTea Editorial · 22 September 2026 at 7:24

Reported by The Independent · How we write briefs