Semafor Sep 22, 2026
With Sasha Baker, Head of National Security Policy at OpenAI
Sasha Baker said a coordinated slowdown by American AI labs runs into China first, because a pause "only works if everybody does it."
The interviewer framed a slowdown as a trade-off against the race with Chinese models. Baker said the goal is not a particular speed but keeping research within what the labs can understand, and that nobody knows yet whether Chinese labs would slow down along with American ones.
"China is the elephant in the room"
Sasha Baker, head of national security policy at OpenAI and previously the Pentagon's acting under secretary of defense for policy, gave the answer on Semafor, in an interview at Semafor's The Next 3 Billion event.
The full segment is covered here so you can skip it.
Here are the 3 arguments that matter.
Key Takeaways
A slowdown only works if Chinese labs slow down too
She describes the goal as pacing, not speed: research should not outrun the labs' ability to understand it
One view holds that China copies American models, so an American slowdown would slow China by default
She said that may be true, and that nobody knows
Chinese labs have tried to match what the leading American labs release, from models to how a lab talks about its security
1. Pacing, Not Speed
The interviewer noted that three AI companies had said in recent days that they wanted some kind of coordinated slowdown or pause while people grapple with the dangers of their models. She asked how a slowdown works if Chinese models are catching up.
She frames it as pacing the frontier
Baker said the term she prefers is "pacing the frontier," and that "it's not necessarily about fast or slow."
The test is whether research outruns understanding
She said the aim is to make sure research is not getting ahead of "our ability to understand it," and that the models stay aligned in the way the labs expect.
