An AI safety pledge needs a way to stop a launch
A reported voluntary White House accord asks AI companies to check their systems and let outsiders test them. Zvi Mowshowitz argues that the promise matters only if reviewers can see bad results and leaders can pause work to fix them. A written pledge is not proof that these steps happen.
What the source reports
Zvi Mowshowitz reproduces and examines a reported White House accord on AI safety. It is described as voluntary. The text calls for companies to watch for risky abilities and behavior in their AI systems. It also calls for staff with power to fix problems, outside tests, and oversight by a board committee. Mowshowitz asks whether outside reviewers will get enough access and whether bad findings will lead to a pause and a fix. His views about politics and competition are commentary. The text of a promise cannot show that any company has carried it out. Nor does this account establish that the accord is legally binding.
Original source
- Title
- A ‘Morally Binding’ White House Accord on AI Safety
- Author
- Zvi Mowshowitz
- Publication
- Don't Worry About the Vase
- Date
- Wednesday, September 30, 2026