Can AI solve AI safety if we cannot check its work?
After a conference, Zvi Mowshowitz argues that we should not hand future AI safety problems to today's AI without a way to check its work. He favors a slower pace. His risk estimates and timelines are personal views, not measured facts about the wider world.
What the source reports
Zvi Mowshowitz writes about a conference and the choices ahead for AI safety. He argues against trusting today's AI to solve future safety problems unless people can check the result. He also calls for a slower pace. The essay describes private talks held under rules that limit who can be named. It offers his own risk estimates and forecasts. Those numbers are his views. They are not a survey of experts or a proven timeline. The essay argues that a proposed solution needs a way for people to check it.
Original source
- Title
- The Curve Bends You
- Author
- Zvi Mowshowitz
- Publication
- Don't Worry About the Vase
- Date
- Wednesday, October 7, 2026