AI Safety Experts Form New Team to Stop Superintelligent AI From Destroying Us

Five top AI safety researchers — Jeremy Gillen, Abram Demski, Sam Eisenstat, Scott Garrabrant, and Kaarel Hänni — have joined forces at a new lab called Resolution. Their job: build a deeper theory of how intelligent minds work. Today, engineers can predict how bridges and airplanes will behave in unexpected situations. But with AI, nobody can really predict what a superintelligent system will do after it has learned, changed itself, and talked with other AIs. The team wants to fix that before it's too late.

The researchers are not trying to make AI smarter. They are trying to understand it. The stakes, in their words, are nothing less than avoiding "the extinction of all life on earth." They see their work as basic research — the kind of deep thinking about concepts, trust, and game theory that could one day help humanity build AI that boosts human freedom instead of replacing us.

Interestingly, the team says even this safety work probably won't help if superintelligent AI arrives very soon. They think slowing down the race toward AI is actually more urgent. But since they can't stop that race alone, they plan to "chip away" at the knowledge needed for alignment while using AI tools to speed up their own research. They admit this might not work, but they think it's worth trying.

The big picture: this is not about a new chatbot or gadget. It's about people who study the worst-case scenarios of AI. They believe humanity is building something it doesn't fully understand — and they're trying to write the instruction manual before the engine starts.