“Recursive self-improvement is the jet fuel that could set AI on entirely the wrong path. And we know these companies are headed that direction. And if we just say that is off limits, then we can start getting on the right path.”
On with Kara Swisher · Tech Strategy & Big Tech · November 2026
Context
An outside expert asked whether Congress would vote to ban recursive self-improvement, meaning AI building the next generation of AI without humans in the loop. Casar answers yes outright.
Transcript
company held personally liable if those processes are kicked off recklessly. The same way we hold executives at meat manufacturing plants or financial auditors personally responsible if they do something truly reckless.
Great question. Greg, then Sam?
My answer tonight is yes. And that to me, RSI, recursive self-improvement, is the jet fuel that could set AI on entirely the wrong path. And we know these companies are headed that direction. And if we just say that is off limits, then we can start getting on the right path. Senator Sanders and my bill to ban artificial superintelligence, one of the ways that we stop ourselves from having an AI that's too powerful to control. An element in that bill is to. Not have recursive self-improvement, make sure humans are building the AI Regulation than having AI that we can't control, building even more dangerous AI that we can't control.
Sam?
Yeah, I certainly support it, but and here's the big but. I don't think we really understand how to implement effectively a ban like that. And I think there are a lot of smart folks, certainly my Nick Lewood. We have a lot of AI safety advocates and labs that are working on things like figuring out…
Right. Although we managed to do it with cloning, we managed to do it with nuclear energy, you know, super bugs and things like that. Yeah,
More from On with Kara Swisher
