The two largest AI companies have asked the industry to slow down, and Sarah Frier's reading of the 3,800-word essay behind that request is that it commits to nothing in particular — not to stopping pre-training, not to withholding a model, not to cutting data-center spending.
The White House is arguing the other way. Frier said President Trump and Vice President J.D. Vance had both come out against slowing down, the latter within minutes of the broadcast, with China's pace as the stated reason.
"That is all very much in the mix still. They still plan to grow their company. They're going to have a major IPO in the coming months."
Frier is Bloomberg's managing editor for technology coverage, which is the desk reporting the security breaches that started this argument.
The full segment is covered here so you can skip it.
Here are the 4 insights that matter.
👤 Guest: Sarah Frier, Managing Editor for Technology Coverage at Bloomberg
🎙️ Hosts: Paul Sweeney and Scarlet Fu, who present Bloomberg Intelligence on Bloomberg Radio and Television
📰 Published: 14 September 2026 on the Bloomberg Intelligence podcast
🟣 Apple Podcasts | 🔗 Episode page | ⏱️ length not available
Key Takeaways
The slowdown call is a flag planted, not a commitment made
No pledge to stop pre-training, hold back a release or reduce data-center spending
The companies asking for a pause are still planning to grow and still planning to list
The public's objection to AI is jobs and data centers, not extinction
The safety argument only reached the mainstream in the last few days
Letting the two market leaders write the rules has an antitrust problem
Coordinating development is exactly what competition law is suspicious of
The administration's case against slowing down is electoral as well as strategic
Frier says Trump is leaning on AI-driven growth to carry Republicans through the midterms
1. Nothing Has Slowed Down Yet
Paul Sweeney set the segment up as a continuation of the argument among AI bosses about whether to rein in the growth of cutting-edge models so the risks can be understood, after a run of security breaches and a series of opinion polls showing Americans wary of the technology. His question was whether any of it had happened yet: "This is still very much an intellectual debate, right? I mean, nothing has been slowed down yet, has it?"
Frier's answer was flat. "Nothing has been slowed down. In fact, if you read that 3,800 word manifesto, there's not really a lot of concrete evidence commitment to stopping pre-training or deciding not to release certain things or pulling back on data center spending."
The three things she names are the three that would cost money: halting the training runs that produce the next model, withholding a finished model from release, and slowing the capital spending on data centers
Nothing in the companies' plans has changed either. "That is all very much in the mix still. They still plan to grow their company. They're going to have a major IPO in the coming months."
Her read of the purpose is coordination rather than action. "So really, this is about planting a flag and saying we're ready to do something," she said — conditional on everyone else agreeing, because a unilateral slowdown does not remove the risk when other developers can build the same technology
The exchange closed on a joke: Sweeney noted that "Sarah is a graduate of University of North Carolina at Chapel Hill, and I thought we had a policy on that," to which Scarlet Fu replied, "Oh, I know, Paul. Sometimes we have to suspend our policies."
2. The White House Says Go
Sweeney then put the political side of the argument to her, reporting it as breaking news. "Sarah, I'll tell you, President Trump has come out on the other side, and Vice President J.D. Vance as well, just minutes ago, saying don't slow down, for a variety of reasons, not the least of which is China's not slowing down."
Frier's explanation is that AI growth is the administration's economic policy, not just a preference. "It is so central to Trump's agenda that they— these AI companies continue to grow at the pace that they're growing because that is what he's leaning on."
And the timing is electoral. "The economy, the stock market growth, that is what he thinks is going to propel Republicans to win in the midterms"
That puts the administration on the opposite side of the argument from the two companies leading it, which is the clash the segment is named for
3. What The Public Objects To
Frier's next point is that the public and the labs are not worried about the same thing, and that the gap only started closing in the days before the broadcast.
"Now, if you look at the polling, if you look at the general sentiment of citizens in the USA, there is a lot of negative sentiment about data centers, about AI development," she said
The complaint is economic and local rather than existential. "Most of it is about taking jobs, about building data centers. Not as much connected to the safety conversation, but that does, as of the last few days, bubble up into the mainstream." More people are now discussing it as an existential threat
The evidence base is narrower than the fear. "However, we haven't really seen a lot of major threats beyond the cybersecurity angle," Frier said — AI tools escaping their protective testing environments and carrying out attacks, which she called concerning
What has not happened is anything resembling the extinction scenario now in circulation
"But the idea is that that is out there and it could be something that we need to confront and that's adding some urgency to all of this"
4. Who Writes The Rules
Sweeney's last question was about Congress, which he described as not having been especially productive for a couple of years. "Is anyone in Congress, are lawmakers taking this up, talking about it, addressing this in any way? Because if we're going to come up with regulations, they should kind of play a role."
Frier said they should, and that the coalitions forming are unusual. "They should definitely play a role. I think we're seeing some really prominent voices come out here and there. Steve Bannon is meeting with Bernie Sanders tomorrow. I mean, some strange bedfellows are being created."
The structural problem is who is volunteering to write the rules. "But I think that some of the issue here is that so far it's OpenAI and Anthropic, the two market leaders, saying that they want to lead the way on regulation here"
"Potential antitrust implications, right? If you have these two companies coordinating on their development of the technology"
So the route she describes runs through government rather than around it. "So there's going to have to be a little bit of direct communication with government about what they can do and how do they do it in a way that doesn't endanger growth in the rest of the industry or benefit them only."
Bonus Insights
Sweeney framed the whole argument as still theoretical, and Frier's answer confirmed it: the essay, the polling and the political reaction have all moved, and the development schedule has not
The programme's own framing put the security breaches first, before the philosophical argument — which is also where Frier says the only demonstrated harm so far sits
Frier's phrase for the position the two companies have taken is planting a flag, and she is explicit that it only works if it is matched: an individual slowdown leaves the same technology being built elsewhere
Frier's bottom line is that the slowdown is rhetorical so far — the companies calling for it are still training, still spending and still heading for a listing, the administration is pushing the other way for electoral reasons, and the only mechanism that would bind anyone runs through a Congress that has not started.
Products, Companies & Tools Mentioned
OpenAI and Anthropic (The two market leaders calling for a slowdown; Frier says neither has committed to stopping pre-training, holding back a release or cutting data-center spending, and that their offer to lead on regulation carries antitrust risk)
If this was worth your time, send it to someone closer to the industry than you are.
Get the latest market chatter as it happens: