Closing Bell: Overtime · Sep 29, 2026
With Bradley Tusk, Founder and CEO of Tusk Ventures
OpenAI keeps saying its own products "went rogue," Bradley Tusk said, and a company that cannot control what it built cannot credibly police itself.
The White House had just signed an AI accord built on self-policing and a "superintelligence czar." Tusk, whose venture firm invests across the AI industry rather than in the big model makers themselves, said the industry's own conduct undercuts the case for trusting it to regulate itself.
"So when you fundamentally can't control your own products, how do you effectively police yourself?" — Bradley Tusk
Tusk Ventures has backed companies including Bird and Latch, and Tusk has spent years advising governments and companies on regulatory strategy around ride-sharing, gambling and now AI. He told CNBC's Closing Bell: Overtime the industry has a narrow window to build public trust before a serious accident forces much harsher rules on it.
The full segment is covered here so you can skip it.
Here are the 5 takeaways that matter.
Key Takeaways
AI companies cannot credibly self-police, because they keep admitting their own products "went rogue"
CEOs who called for regulation two weeks ago want a level playing field, not government oversight for its own sake
A public trust problem, not a regulatory one, is the real risk to the AI buildout
A poll he cited found people prefer a prison in their community to a data center
OpenAI's decision to hold back its October model wasn't voluntary, he said, but a response to weeks of bad headlines
The math on an OpenAI IPO is hard, given how much the company would need to grow against a market already skeptical of AI
1. Self-Policing Won't Hold
The host asked Tusk whether self-policing, a White House accord and a "superintelligence czar" were enough to manage AI's risks.
Self-policing only works if companies actually control their own products
"Probably not. Simply because if you could believe that the companies and the frontier models all have full control over their own work, and then yes, self-policing arguably could work." — Bradley Tusk
"We keep seeing every day now, at least out of OpenAI. Oh, this thing went rogue. That thing went rogue. We're sorry this happened. We should have disclosed that." — Bradley Tusk
The host noted that Anthropic's Dario Amodei had broken with the White House's own framing at the summit, telling the room the industry still had to "pursue this in a safe manner" rather than agreeing nothing more needed to be done.
Regulation is really a request for a level playing field, not for oversight
Tusk said the same AI chief executives who called for regulation only two weeks earlier were not doing so out of faith in government.
"You saw about two weeks ago, almost all of those same CEOs say we should be regulated." — Bradley Tusk
"They want something that sort of creates an equalizer for each other to say, okay, we can only go this far and we don't have to worry about limiting ourselves in any way because everyone else is limited to, and therefore it's a fair playing field." — Bradley Tusk
A serious accident would bring regulation ten times harsher than today's
"I think the real risk they're taking politically is if there is something really hugely tragic, right? An air traffic, the air traffic control system goes down, a nuclear power plant, whatever it might be, then the regulations will be ten times worse than what they could have right now." — Bradley Tusk
