Trump Admin Curbing Chinese Remote Chip Access, 100+ Firms Want Better AI Cyber Defense Policy

0:13Welcome, everyone, to The Information's TITV. My name is Leo Schwartz, and I'll be filling in for Akash today. It is Friday, August 28th. Today on the show, the information has exclusive reporting that the Trump administration is working on new rules that would limit China's remote access to chips. We'll then have more details surrounding Anthropik's highly anticipated blockbuster IPO expected later this year and how it might be different from other big tech listings this year. We'll then turn to data centers, which have become one of the most hotly debated issues ahead of this year's midterm elections.

0:45We'll look at both sides of the policy debate and what it could mean for future regulations. And to close out the show, 116 companies signed a letter calling for businesses and policymakers to prioritize cybersecurity in the age of AI. We've got a special edition of the editor's cut where we'll break it all down. It's going to be a great show, so let's get right into it. This morning, The Information published exclusive reporting that the Trump administration is working on a rule that could cut off a key loophole allowing Chinese AI firms to access advanced U.S. chips. To talk about the finer points of export controls, we have Michelle Nye, a visiting fellow at the Center for a New American Security, who is one of the experts featured in our article.

1:28Michelle, let's start with the basics here. What has been the Trump administration's approach to handling export controls of AI chips to China? Hi, Leo. Thanks for having me on. The Trump administration rightly needs to weigh many different considerations from commercial interests and economic interests to national security risks. So their approach in expert controls has really been to strike a balance between all of those. And what did they change from the Biden administration? Yeah, the Biden administration was more heavy-handed on expert controls. They really valued the national security considerations of exporting chips to China, and they were more willing to export control of those chips.

2:23And I think the Trump administration has been more open to weighing a wide host of different considerations. So as you reported, the Biden era saw the AI diffusion rule, which was a pretty comprehensive framework that determined which countries in the world could receive how many chips. And the Trump administration has been quite vocal in criticizing this framework and wanting to rescind it and replace it with a new one. So in the wake of that replacement, which we reported about today, critics argue there are loopholes in the current approach. Remote access, I know, is chief among them. Can you just break down what the remote access loophole is and why it matters?

3:08Sure. So Chinese companies are accessing large swaths of compute remotely, and this is compute in the form of NVIDIA advanced chips such as Blackwells that they would otherwise not be able to buy outright themselves, but they're able to access through cloud computing providers, most notably in Southeast Asia, but also cases have popped up where they're customers of American cloud providers as well and accessing that compute. And they're using this compute to train often very advanced models. For example, Michael Kratzios, the director of the Office of Science and Technology Policy, recently said that Moonshot AI used remote access to train its latest Kimi K3 model.

4:00So we're seeing quite advanced models being used, yeah, using remote access to train these innovations. So why has this not been closed yet? What are the challenges of closing the remote access loophole? Does commerce even have the authority to do this? So there are debates over whether commerce has existing authority to regulate remote access. The Bureau of Industry and Security has not interpreted its authority to extend to cloud services, and they've expressed this in several advisory opinions. So it's unclear whether BIS or Commerce currently has the authority to regulate remote access. But either way, this authority should be clarified in order for BIS to actually make rules about it.

4:58So we reported in our story this morning that there's a small group within the Commerce Department that's working on a replacement to that Biden-era diffusion rule that you mentioned. You talked about this a bit, but what have been the ups and downs of the Trump administration's approach to the diffusion rule, actually rescinding the Biden one, coming up with a new one? What's been happening there? So the Trump administration has not formally rescinded the AI diffusion rule. They announced in May 2025 that they would be rescinding it. And shortly after, BIS announced that it would not be enforcing the diffusion rule.

5:33Now, this approach has created some national security risks in that it left open a loophole through which Chinese subsidiaries abroad were able to purchase advanced NVIDIA chips. So that loophole was closed in May 2026. However, in the meantime, experts estimate that potentially hundreds of thousands of advanced chips were bought by Chinese subsidiaries abroad. So we reported this morning that if there is a new diffusion rule that's being worked on, it's likely to address that issue of remote access. Obviously, you mentioned a different loophole that was previously closed. If remote access is addressed in a new rule, what would the impact of that be to Chinese AI companies and U.S.

6:23chip manufacturers? Yeah, so as I mentioned, Chinese companies are getting a large amount of their compute remotely through cloud computing providers. And they are using that to train their most advanced models. So if they didn't have access to this compute, they would need to find other methods or ways of training and accessing compute. And so that would definitely slow down their AI ambitions. And then what about for U.S. companies? Because I know with the previous diffusion rule under Biden, there was a lot of pushback from U.S. companies as well. There was accusations that it could hurt U.S.

7:03competitiveness of semiconductor manufacturing or other similar companies. the U.S. Yeah, that has definitely been the pushback from industry. There's also been pushback from allies as well that weren't included in the tier one of that AI diffusion rule framework that those countries were able to get relatively unlimited access to chips. So it's clear that there were some flaws in the old AI diffusion rule under the Biden era. However, this diffusion rule did also include a stipulation to control remote access, which it seems like the Trump administration replacement is also considering. So that will be critical to protect national security.

7:53Given the track record of the current Bureau of Industry and Security and the Commerce Department, what do you think the likelihood is that they'll actually advance new rules? I think their ability to advance new rules without some sort of clarified authority, either by this AI diffusion rule replacement or something like the Remote Access Security Act, which would clarify BIS's authority to regulate remote access. I think without one of those actions clarifying their authority, They might be unlikely to regulate remote access on their own. I think having at least one of those in place would definitely make it more likely that they would regulate remote access.

8:40What's happening with the congressional effort that would actually give Congress authority to do that? Yeah, so Congress is weighing the Remote Access Security Act, which is a bipartisan bill that would regulate remote access and extend export controls to remote access. It has currently passed the House of Representatives. It's still awaiting action by the Senate. But if passed, this would give BIS the authority to regulate remote access. because now BIS would still have to create a rule to actually export control remote access, but their authority to do so would be clarified. Yeah. I mean, do you think, obviously, given the fact that the previous iteration under Biden did face serious pushback, I know there was a draft of this that was released in March by the Trump administration that got a lot of pushback.

9:31What do you think would happen with a new iteration of this? What type of pushback might it receive? Yeah, any new rule is going to have pushback from industry, from countries around the world. I think the Trump administration has signaled that it would want to do a more inclusive approach to any replacement to the diffusion rule. Which, yeah, I think would definitely make allies and partners around the world that perhaps weren't included in tier one the first time. Yeah, they would be more supportive of the AI diffusion rule. But I think anytime you try to implement expert controls, you will get pushback from industry that cares more about commercial interests.

10:21But it's really the job of the Trump administration to weigh those commercial interests against national security risks. And then assuming that remote access is addressed in some way in these new rules, what are the other most glaring issues that you see that the Trump administration still has to address when it comes to AI export controls? Yeah, remote access is definitely a big one. I actually have a report coming out soon about remote access and all other legal gray areas in which Chinese companies are accessing American compute. There's other vectors of access, such as Chinese companies building data centers kind of wholesale on Malaysian soil or in Southeast Asia, and those are much harder to track.

11:11A lot of these Chinese companies also obscure their identity when buying chips or building data centers using shell companies and complex financial arrangements. So I think being able to investigate and understand those incidents and monitor and stop them from happening will be very crucial in export controlling advanced chips. Well, I'm looking forward to that report. Thank you so much for joining us. That's Michelle Nee, Visiting Fellow at the Center for a New American Security. Thank you so much. We have some new details on Anthropix's upcoming Blockbuster IPO. The company is working on a plan that would let existing shareholders sell some stock during the IPO.

11:54Our colleagues Valida Pau and Corey Weinberg wrote this story. Corey joins us now. Corey, what do we know so far about Anthropix IPO? Do we know when it's happening? Yeah, it's coming up. We don't know the exact dates. You know, that won't be released for a little bit. But here's what we know about the timeline. We know that sort of what's going to happen next is Anthropix going to publicly flip or publicly unveil its IPO prospectus. That will come sometime in September, definitely after Labor Day. These things can always face hangups, but we expect it to be sometime in the first couple of weeks after Labor Day, which would put it in line for an early October IPO.

12:42well i would say that i i hope you're getting some rest before labor day and eagerly checking your phone at every notification but i know you've already been reporting closely on what's happening before then i mean the first is this story that you had yesterday with valida what did you learn about how anthropic seems to be treating existing shareholders for this ipo anthropic is taking a different tact than what spacex did earlier this year and what Reapers did earlier this year with their very large IPOs. And that Anthropic is, at least as it stands now, planning to allow existing shareholders to sell at least some stock in the deal.

13:21This isn't a complete anomaly. This does happen for various reasons. But traditionally, we think of an IPO as this event that raises cash for the company. And we think of it as an event that gets a company public, which will eventually allow its existing shareholders to sell. Because we have generally this norm and this expectation around stock lockups. You can't sell your stock if you're an existing shareholder for a while, sort of after your IPO. And Dropik, at least as it stands now, seems to be taking an approach where they'll allow existing shareholders to sell some stock in the deal. That would put the stock in sort of the hands of public investors that Anthropic would be able to choose.

14:12And then shareholders would potentially be locked up, not able to sell their stock for potentially even longer than the 180 days that people are generally locked up for. And it's important to say that, look, these are the kind of like real nitty gritty structural issues that a company is sort of figuring out and deciding ahead of their IPO. And this is obviously a really big IPO and like pretty much the entire, you know, sort of market economy rides on Anthropics sort of stock market success right now. And so I think they are doing a lot of deliberations. And this is kind of a structural element that we've heard is sort of likely to come to pass.

14:55But, you know, they're still hammering things out. I mean, is that a bearish signal if you have early investors and executives and employees of Anthropic who are trying to offload their shares immediately? This seems like a very ideologically driven company. You'd think they'd want to hold on to their shares if they think this company is a rocket ship going to the moon. 100%. That's like exactly the sensitive issue we were sort of thinking about and trying to get more information around ahead of the story. I think from talking to lawyers and other types of advisors who have worked on a lot of these IPOs, the bankers generally do tell any company, look, you don't want to be selling too much stock in an IPO.

15:40It'll look like to everyone like you're dumping your shares before the stock has sort of reached its full potential value. um because remember like an ipo is generally priced you like you are generally selling shares at a price that everyone sort of acknowledges is supposed to be uh sort of a discount for the buyer like you're not selling at a price that that anyone would acknowledge is supposed to be sort of the best price to sell your stock at so yeah it can be a little weird but I think there's some tactical reasons why people may need to do it. They may have a potential tax bill when their stock vests, and so they would need to have a decent amount of cash for that.

16:27It could be for Anthropic to control how much stock they can put into the hands of new public shareholders or which shareholders they give stock to. And then it could be because they want to do this other thing, which is have an extended lockup period to ensure that the stock price isn't as volatile as we've seen with SpaceX and Cerebras. And I think some anthropic shareholders have to buy galaxies from their wives if public reporting is to be believed. Yeah, I mean, they need the cash for that. The Martians are not just handing over the other galaxies for a measly billion. yeah still still curious how that process works of buying a galaxy but i guess that's a story for another time um there's something else that you reported in your story which is a seemingly contradictory policy for how anthropic is considering letting employees sell stocks after it goes public can you just break down what's happening there yeah that's what i'm talking about with the extended lock-up period and it seems contradictory but they actually would go hand in hand um because the reason why you one of the reasons why you potentially would let existing stockholders sell in the deal is to give them sort of an opportunity to get some liquidity quickly but need to hang on to their stock for for longer than is is typical um generally with a lot with a with a big stock unlock um for a company like spacex or a company like anthropic where, you know, essentially when you, the thing to remember about an IPO is when you go public, you know, because of this lockup period, because existing shareholders generally can't sell, you're only floating, you're only making public a relatively small portion of the actual outstanding shares in the company.

18:20And so that just creates, you know, it can create sort of stock price volatility because there's just not that much stock in the hands of public shareholders that could actually trade. And then what creates a lot of volatility and like share price gyrations is this idea of stock becoming unlocked from employees and investors. They're able to sell for the first time. Hedge funds trade around these events, especially like SpaceX and Anthropic, where there's a huge amount of stock that will be coming unlocked. There's not many other trillion-dollar companies in the stock market, and these are two companies that will all of a sudden have potentially an outflow of existing shareholders looking to sell.

19:07And we've seen that with SpaceX. They did some early unlock maneuvers, which has sort of, people think, sort of contributed to its relatively sluggish stock price performance as a public company. What do we know about the latest valuation for Anthropic? Yeah, what we reported in our story is bankers have at least discussed$1.5 trillion in recent weeks. But there was a report out from the New York Times last week that said Anthropic was considering$2 trillion. I think the way I look at it is it's generally too early for anyone to have made decisions on this thing, you know, sort of with enough finality.

19:52Like what's happened so far, the topic has gone, it's filed a test one confidentially. It's IPO prospectus. It did that in early June, late May. In recent weeks, really throughout sort of most of August, it's been meeting with investors, public shareholders, potential people they want to invest in the IPO and giving them a bit of a look under the hood. It's called a testing the waters meeting. And, you know, they're still trying to get feedback on what a public investors want to, you know, see from Anthropic, but what resonates with them, what doesn't. And so they're trying to get a sense of price, but price is not necessarily getting like half out at this stage.

20:38And finally, without giving away too many of your reporting tricks, what are you on the lookout over the next few weeks? What's coming next? Yeah, I mean, look, I think the S1 document will be probably the most highly anticipated IPO perspectives, at least in my career. But probably a lot longer than that. And that will, I think, give us a really good look under the hood about the economics of a big AI lab. I think there's so many moving pieces right now in terms of how do what these companies sell interact with open weight models. There's a lot on sort of what is actually driving customer demand, where customer concentration is.

21:27So these things are going to be picked apart like crazy. and some of the things the company can't control what they disclose or can't control what they disclose because it's accounting rules and really strong norms. But companies do have a decent amount of leeway to decide, like, yeah, this is what we want to tell the world and investors about our business, and this is what we don't. So I think with Anthropic, it is a relatively secretive, it is a very secretive place. You know, it definitely draws comparisons to Apple in terms of corporate secrecy. But there is, you know, this sort of really strong cultural element on top of it.

22:08So I think that will make this IPO perspective even more interesting. Well, I know we'll be able to read it in the information first. Hope you're getting some sleep. That's Corey Weinberg, our senior reporter covering AI and business for the information. I'm excited to be joined by two of the sharpest minds in AI policy who have very different views on the debate over regulation and the role of the state in advancing AI safety. Brad Carson is a former congressman who now leads two AI safety nonprofits, including one of the top AI campaign finance groups. And Adam Thier is a senior fellow at the R Street Institute and writes widely on AI governance and regulation.

22:46Let's start with perhaps the most divisive question in AI policy right now, which is data centers. Adam, I saw your recent critique of Ken Paxton, who's running for Senate in Texas. and argued that data center companies should be held criminally liable for AI chatbots that undermine children's safety. What's happening right now with politicians and data centers? How do you make sense of this? Well, we're in an environment of fear and loathing, if you will, about data centers. And it seems like this is probably the most repressive set of policies I've seen for digital technology in my 35-year career.

23:18With regards to data centers, we're talking about banning them outright or placing moratorium on them. The Ken Paxson proposal includes one of the most radical provisions I've ever seen, which is that it would hold data centers liable for what AI chatbots do downstream. That's a little bit like saying that electricity companies should be held liable for anything that is plugged into them, or a newspaper should be held, or an ink maker or a newspaper should be held liable for something else somebody else writes. I mean, we can't have liability work that way. It'll just destroy the industry and destroy speech.

23:48So hopefully that's not where we're going to end up on data centers. Brad, what do you think is the responsible path for approaching data center regulation? Does one even exist? I think one does exist, and that is that the hyperscalers and the people who work for them have to be more responsive to the communities. Obviously, there's a lot of concern about AI, and I do think that in some ways data centers are a stand-in for a broader concern about the technology. At the same time, there's a lot of opacity, there's massive asymmetries and sophistication between the hyperscalers and, say, a rural county commissioner who's trying to negotiate on it.

24:23They're giving away, in many cases, my home state of Oklahoma, we give away our tax base to recruit a hyperscaler in. I think a lot of people have legitimate questions. So we're going to have to build data centers, no question about that. But I do think there's going to have to be more community involvement, more community transparency, and the hyperscalers are going to have to put up some money to rebuild the schools and the roads, to ensure that the power prices don't go up for the companies. That is a path forward. And I think we'll get that Adam, I know what you're opposed to with data centers.

24:55Do you think that any regulation makes sense? I mean, it's such a hot topic right now. There's a lot of dissent that's going on, but what's the most common sense approach here? Well, obviously there's going to be some regulation of data centers where they're placed and how they're zoned and things like that. That's the same for all industrial buildings of any sort. The real question is, are we going to allow them to be built at all? And the problem is now is that the data centers have become sort of a proxy war for like all these other concerns about AI policy. All of those different concerns can be addressed in various ways, but we should not address them by essentially demonizing and destroying the infrastructure layer of the entire AI staff, or else the entire AI revolution and robotics revolution just cannot happen in this country and it just moves to China and they will leave the globe on it.

25:41We need to take a more focused approach. We can address the various concerns already dressing, obviously, concerns about things like electricity and water and various other issues, but it can't become an excuse to try to address everything under the sun with regards to concerns about AI, or else we'll just never have any sort of computational revolution in this country. Another perpetual flashpoint that may have receded a bit recently, but I think will come back roaring soon, is state versus federal preemption. At this point, it seems we're not going to get anything out of Congress before the midterms.

26:13We're seeing some new state actions, including out of Massachusetts. Brad, should states continue to take action on AI safety and policy? They're going to have to. I do think we've won the fight on federal preemption. It still lingers out there in a certain rear guard way, but most members of Congress aren't for it. Republicans generally don't like this kind of attacks on federalism. Democrats think it's a sop to the industry. So while a few people still push it, I think the issue of state preemption is largely one in our favor. I do think the states are going to have to regulate on it. Federal kind of preemption is a natural part of a federal system.

26:46So if we have a comprehensive frontier law, it makes sense to preempt states from just duplicating that or working at loggerheads with it. AI is a broad sector. It's not just the frontier. It goes down into narrow models that have a lot of application in state and local government and companies. That's appropriate for the states to look at consumer protection, civil rights, labor laws, things like that. So I think the key is there's going to be some kind of preemption in the future. is going to have to come alongside a comprehensive law, probably only on the frontier, while leaving states to regulate what are their historic areas of concern.

27:22Adam, do you think it's a bad idea for states to be taking action here? Well, I'm very troubled by the rising patchwork of different conflicting, confusing, and costly state and local policies. We're talking about somewhere on the order of 1 ,850 different AI related proposals across America right now. And we really have patchworks within patchworks within patchworks. We have states regulating all sorts of interstate activities regarding AI frontier safety, surveillance issues, various types of questions about, you know, pricing online, so-called AI bias and fairness. The reality is that some of these things may be right for state activity, but the reality is also that if we have that many different confusing overlapping laws, at some point, this has massive ramifications for the development of a national marketplace.

28:10This is not how we regulated the internet over the past 25, 30 years. Congress set up a national framework on a bipartisan basis with the Clinton-Gore administration. It worked marvelously for the United States. We developed a world-leading capacity in computation and digital technology and e-commerce. We are seeing a reversal of that, and I'm hugely troubled by it because this is where we need the voice of Congress to sort of equalize things, level the playing field, and make sure we have a clear, consistent policy for the United States and its leadership in global technology. Brad, what's your response to that critique, this idea that we're just going to have a patchwork of different state laws?

28:45Obviously, if you're open AI or anthropic, it's very difficult to deal with 50 very different laws if you're trying to move quickly in building your new models. I think I have several responses to that. The first is the internet was much more monolithic than AI is. As I mentioned, AI is giving me my Spotify recommendations. The next movie to watch on Netflix. Flock is using it in some ways to understand the surveillance. So all the things Adam talks about are open to a massive federal bill. You have something like the EU AI Act on steroids. If they want to cover surveillance and frontier regulation and civil rights and consumer protection and labor law and kid safety at the federal level, you can honestly talk about a preemptive effect.

29:24But really what people who push preemption are to say is like, no, we want no law on those at all. They don't offer a federal standard. They just say like, we'll block the federal government and then let and try to block the states too. So the internet and frontier models can be analogized in some way. That's why you're going to have probably a federal law about frontier regulation. But AI is so comprehensive in our society today that you're going to have to have a lot of state regulation or else the federal government would have to stand up new agencies. There'd have to be a bundle of new laws.