Emergency Pod: Claude Fable Fried + What's Going on at BIS?

abs, my mayor Muslim, my bagel Jewish. Fable is fried. Next in five, congratulations to the New York Nicarbockers in the city of New York on their first championship and 53 years and a hearty perhaps message of concern for the Trump administration's bureau of industry and security and commerce, which has been taking care of America's lead about as well as D.R. and Fox to care of the basketball over the past two weeks to describe why Trump took Fable offline. What's going on with chip AI loop holes and what this all means for the future of the AI development in America and beyond. We're just we're joined today by date of New Yorker and long time state department and White House civil servant covering AI shifts and whatnot. You know, if you love him, Chris McGuire, welcome back to the show. I've been a nix fan much longer than I've been any of those things. So that's the most important thing right now. But yeah, I had to had to wait at your urging had to wear the jersey to celebrate. So we won't be taking it off for well, at least a week or until it smells. So just a brief recap. We recorded a model talk on Thursday or Friday of last week, where we were all talking about our Fable impressions and philanthropics latest model. Now we have a nice little note when you go to claw dot AI saying clawed Fable 5 is currently unavailable. Now, why is that still sort of unclear, but the the facts that we know is that some point on Thursday, Andy Jassy gets on the phone with someone in the Trump administration and says, Hey, there's something fishy about this whole clawed model. We found some jail rates, which can do something to have you get access to some capabilities, which anthropic didn't want the world to have access to. There are some sort of communication between anthropic and various corners of the Trump administration. And then at five twenty one, we get a new export control in which I hand over the reins of this narration to Chris. So what did that? You know, what did commerce put out and what were the implications? Yeah. So there was a is informed letter that was some of the details are still, you know, not publicly reported, but basically there's an is informed letter, so there's a private letter that was sent to anthropic that one way or another regulated seemingly the export of the model, the remote access of the model or the, you know, the ability to access the model via an API from other countries or from foreign nationals inside the United States. So of these, you know, obviously the the commerce department has regulated the export of models before in the Biden administration. The Trump administration actually interestingly is not enforcing those particular export controls, but now is introduced these. The authority to regulate foreign persons is it called a deemed export control. It's something that in the Biden administration, we actually generally exempted most of the emerging tech controls from deemed export controls because of concerns on how it would impact business in the United States. But you can say that a foreign national inside the United States requires a license to receive a good that is controlled for export outside the United States. And then it's a little bit unclear how all this fits together in a way that actually regulates API access. There's probably ways that you could do it, but without seeing the letter, it's a little hard to tell. But the net result is that it was illegal to export the model, use the model by any foreigner either outside or inside the United States, which obviously made it very hard to serve the model, partly even in the United States, because you would have to verify who and users were. Presumably, there are actually probably employees at Anthropic who are foreign nationals who all of a sudden are restricted from accessing the model that they work on. So this obviously kind of put the use of it to a stop. And then now it seems like there's discussions going on today to try to resolve this one way or another. So first off is Andy Jassy, the Jeremy Sochan of this story, Jeremy Sochan, of course, famously were drafted by the Spurs, worked there for three years, got caught, shows up to the next and wins a championship by spilling all of their internal workings and secrets. I don't know. I mean, I don't know what the motivation is. It will be interesting to see on the next side how much of a behind-the-scenes role Sochan played in the championship and how much we had all the all the intel due to a crafty signing mid-year. Or did the next get access to mythos during the glass wing window use it to hack, you know, when he's email, like get all the dirt on him in order to like whisper, you know, disturbing things when he was taking free throws. All right. We'll stop. It would not be the first time it sports team has hacked another one for competitive advantage, but yeah, I don't think it happened here. All right. So clearly there is an overlay of like bad blood or at least no love lost between this administration and Anthropic. But the sort of more remarkable piece to this story from my perspective is you had this administration talking for two years about how, you know, they wanted capitalism to run free that like AI shouldn't be regulated. This is this is something that was like important to American development leadership and and sort of those like those safety knicks in the Biden administration really were we're overreacting. And then, you know, you go from that to having a having an executive order, which like was going to be, you know, not all that tight be watered down even more when it came out two weeks ago to all of a sudden over the course of apparently 24 hours, them deciding that the most powerful model in the world, which is like a, you know, instantiation of America's lead over the rest of the world needs to no longer have access, users can no longer have access to. So Chris, curious for your reflections on on that ARF could maybe what it tells you about the future of AI policy. Yeah. I mean, look, the the particulars of this individual incident aside, the administration very clearly has fundamentally changed its policy on AI over the last few months. I think that mythos, the mythos released really did kind of cause a sea change in Washington on AI policy. The policy has not been, you know, let it rip, let the private sector rip for, you know, four months or so. I think you saw the culmination of that in the AI executive order that, you know, put forward a transit voluntary licensing regime, but, but ultimately are a voluntary safety regime. But ultimately, it's obviously kind of mandatory. I think this basically shows this mandatory. So it's, it's logical that the admit, this is a real problem. And I think that the administration is realizing that this is this meeting like AI regulation and the very powerful capabilities that these models have requires regulation. And to the administration's credit, I think they recognize that and they have pivoted at least in a broad philosophical sense. They haven't pivoted all elements of their policy and they also are moving in ways that are very reactive and fairly chaotic. And I think what's clearly needed is you need a policy that is, first of all, not just reactive that actually there is a kind of framework that's set out of like here's the expectations. Here's how we regulate things domestically. Here's how we regulate them internationally. Here's what people can do. Here's what people can't do and has a degree of durability. I say a degree because like, look, this is regulating at the front here and it's hard. And we did this in Biden administration and you know, there's degrees to which we would have to update the controls and then business would yell at us and say like, well, but this is not predictable. And we like, we're doing our best. Like we're trying to create as predictable as environment as we can while responding to national security risks. But like we did have some degree of a framework out there. And you know, the administration needs to do that. They need to, they need to be clear for like, what is and isn't allowable domestically? And like what the, what, you know, there's, there's apparently some standards in the AIO for what would, you know, how they would, they would do model evaluation, but they're classified probably helpful if those were public. But regardless, you could work with it. There's not, there's not that many companies. You also need a corresponding strategy internationally of which way I think we have none of how are you going to ensure AI infrastructure globally is safe? And how are you going to make sure that other countries, when they start making this level of capability, which they do, which they will very soon, are not, you know, used against us in the ways that people, you know, the administration is correctly worried that people could inadvertently use, you know, current generation models against this too. So like there needs to be a comprehensive strategy here. We're just not seeing that. And I feel like it's like a trope in Washington. I'm like, oh, we just need a strategy, right? It's all I've run says to anything. So I'm almost hesitate to say that. But like, look, we export controls in particular are a powerful tool. They're a useful tool. They're a tool that can be used to actually ensure that technology is given to the people you want it to be and to create an international non-proliferation regime to a degree, but we're not doing that at all and instead we have this insane situation where right now Canadians are prevented from accessing these US models, but meanwhile the chips that make the models are free to be sent to many of them to China because of loopholes that the administration created and also to all other countries around the world and there's no efforts to kind of get the global regime in order. So look, I hope that the administration recognizes that AI regulation holistically is important. Also this is an indicator, export controls are an important and powerful tool and something we should use and I hope we actually start moving towards a smart export control strategy too while also having a domestic regulatory strategy that is logical and predictable and that has good relationships between the companies and the government where it doesn't make people feel like the $700 billion in CAPEX, the hyper-scalers are spending are pointless. You do need to make people feel like that investment is going to pay off. But I'm kind of concerned that this chaotic environment right now is even though these are real policy problems, the administration is trying to address in terms of fast evolving national security risks in AI, it's doing it in a pretty ad hoc way and the risks of that are just tremendous. Yeah, I was talking about some of those risks. So one, I mean, you mentioned it like popping the AI bubble, right? Yeah. Like if the conclusion stocks are up today because the Iran were ostensibly ended, but you know, if we end up in a situation where models can't get any better than they are as of, you know, I mean, not even June 2026, but say like March 2026, like a lot of that CAPEX is kind of for not because this is sort of like baked into everyone's projections is that like the models keep having increasing levels of economic utility. So clearly, I don't think that's what they're necessarily aiming at here and I for one have some sort of load star of like even with this action, Trump not wanting to put on us like a regulatory system, which ends up like, you know, appreciably squeezing American GP growth. But I think that is like the one that, you know, people people like are talking about, oh, what's the, what's the economic impact of like AI export controls China like does the video make $50 billion more or less? I mean, having models not get any better in America or having there be like significant uncertainty, that sort of models can improve at whatever rate the technology would allow them over the next few years is a would be a real problem to this entire, this entire build out we can. All right. Let's let's go back and forth. What's your? Yeah. What's your risk of this going on? No, that's exactly right. And look, there's, you know, three companies right now that are relevant in the frontier space. And, you know, how much that's not it, it's good that we have some competition. It's not a ton. Like how much there's a real question of how much can the market bear if, you know, one of them, two of them, certainly all three of them are told like, sorry, you just can't release anything that's, you know, more capable or even not that, right? Because presumably the government does want to work this out. But it's just if the perception from industry is that like, even, you know, even if, like, there's no way to release a model without being subject to some kind of insane draconian measure in response that, you know, very rapidly without an effort to work it out that then causes like cascading effects, like at what point does that create actual like business investment problems? Because I don't know and some of this might just be, you know, there obviously is, you know, there's animosity here between the government and the company. So maybe this just isn't something that would ever happen with open AI or with Google. But like there are, you know, real national security risks here. And I think there's going to be issues where whether or not this one was or not, where, you know, the government has to work with the company to rapidly fix problems. And they need to find ways to collaboratively do it together and just saying like, nope, sorry. We're just going to impose like completely draconian controls that shut down the company, you know, very quickly. I mean, if the company was actually acting in a way that was like very actively hostile, then obviously yes, that's warranted. But if it's not, then like it's really in everyone's interest to work together to try to resolve this as quickly as possible because otherwise the spillover effects are going to be really big. And I'll bet you all the other companies are probably thinking like, well, what if what if this does happen to us like, I don't know, should we, if we create a model that's as, you know, as powerful and as concerning, are we just inevitably going to get hit with and is informed letter once someone finds some issue somewhere sometime? It's a hard, it's a hard, it's a hard dilemma. Here's another list that I feel like has not really been explored. So there's the issue of export controls, obviously on, on kind of China and other countries. And whether or not we should do that for their own purpose. Obviously, I think that we should and we should be trying to hold them back. But like part of the reason that this really shows why you want to be holding them back, because the domestic regulation is hard. I will acknowledge that this is a difficult problem and that like you do have, you know, saying all the administration needs to do is just develop like a comprehensive AI regulatory regime is like, kind of unfair and overly simplistic, because this is a really, really difficult thing, right? Like it's building this in a way that is smart and robust, but predictable, like, look, the bi administration, we spent a year talking about the AI diffusion rule and that's just one element of the problem, that's just the international element of the problem. Like just saying, like, just, just, just go figure this out, you know, it's hard. So the point here is like you actually want to have a really big lead over China because you're going to need time to figure this stuff out. And if you miss fire on some of the regulatory measures, you know, it will, it will slow you down. And you want to have the time to make sure that like you can actually be deliberative, think about it, you know, come up with the best kind of structured regime that is actually going to work and not blow the lead to China. And the more the farther you are ahead of China, the more you are actually able to do that. But if we're neck and neck with China because we're actually not doing anything on international controls, you're going to be forced into the environment that we're in now where you have this kind of ad hoc chaotic regulatory regime is going to produce regular worst regulations. It's going to have more incidents like this. It's going to have a spillover impact on business and investment. And it's ultimately going to be worse for everyone. So like having that time actually creates the more predictable environment, the better regulations at home and that's better for everyone. That's actually one of the key underlying points of all the controls and it's a sympathetic point to government because this is really difficult. But like, you need to give yourself time to do this. And if you don't do that, then you're also at fault for not not giving yourself the time to do a hard problem. You just you end up, yeah, okay, you, you know, you did all your homework the night before and your paper was mediocre. But like, maybe you shouldn't have started the night before, right? Give yourself actual time to do a good job. Yeah, I mean, there is an aspect of this where I'll give some grace to the administration and that like, okay, what's our analogy like I tar for fighter jets or something? Like you know what a fighter jet can do years before you decide who you want to sell it to. Like it has this range, this speed, this, you know, EW system, these types of missiles, right? That there is an aspect of these to these models where, you know, whatever creative benchmarking and stress testing and red teaming, like 10 or 100 or 1000 people can do, it's just an entirely different beast once you kind of like let it, you know, give it increase the increase the circle of people who have access to it and I think like the idea that you wouldn't be, like the idea that folks wouldn't be able to get around these sorts of controls, particularly which are new. I mean, I remember like, I guess the question to me is like, is the challenge that anthropic is facing currently of not letting actors, they don't want to not letting just like a random user be able to, you know, improve their AI model or generate hacks from creative prompt testing, something that like is ultimately like a stable and fixable one, like Google had at the early days of itch image generation where people make, we're making those black Nazis and they just like thought it was ridiculous, or is there something sort of like like deeply structural which sort of more time and research is not going to be able to necessarily figure out that these powerful models, like if you give, if you give a wide enough sort of like access surface to them, folks are going to be able to do, you know, basically whatever they want to. It feels like it's the former, I think it's the former, yeah, I mean, it's a fair question and when we need to, you know, carefully think about it, we've gotten better over time, right? Like it's harder to jailbreak models now than it was. year or two ago. So that seems to indicate that like, you know, this is something that we can continually approve on, especially with admins and AI. Now, will it be ever like a completely zeroed out risk? Like, that's a little hard to say. Um, but you can certainly probably get this risk very, very low. And that's also an area where like good collaboration between industry and government would be really helpful. Like if you have, you know, NSA and, and Casey, you know, really doing robust stress tests in cooperation with industry on the models to make sure that, you know, there is jailbreak proof as possible. Like that is good. And then if there's something else that comes up, then you work immediately to correct it. Um, but like, I don't know, I, I, I feel like that is as long as you can get it to a very, very low level. Um, I don't know. I feel like we have reasons to believe that that's possible. But they're, you know, technical people who could have more, you know, robust thoughts on that than me. But like, that, that, it's not something because the alternative is like, we basically just have to, have to either live with like some crazy risk or just stop development, which also isn't going to work. Because eventually you're going to, other people are going to get to that level too. So, um, yeah, you just want to invest in that, that R&D to, to, you know, make this better. But you, you have to, you have to really incentivize it, right? And it has to be collaborative. Uh, I'm just thinking like the, the, the nicks and their coach, like they, like this profession, right? Um, uh, you are pulling from, I don't know, a hundred million people who've like dribbled a basketball in your life. And there's such a kind of effective professional win-owing process that like the top folks are compensated extraordinarily well for, um, really, um, you know, really remarkable elite performance. And then we have Casey, which like, look, no, no disrespect, but like your budget is less than, you know, two weeks of Jalen Brunson's time. And there's something not right with it. Like it's a big deal. I don't, yeah. It's like, well, Landry Shamit, right? The Casey's like on a non-guaranteed contract right now, right? Coming into training camp and like then they're just hitting three after three after three, right? Like, man, maybe this guy should actually get some, some, get paid. Uh, yeah, or just like not get like, like their, their top hire fired because, you know, their former employer is like, you know, on the, in the dog house or something. Anyways, so I think in parallel to this, we started seeing all these blog posts about, uh, RSI sort of like recursive self-improvement. And models that have like powerful side alert capabilities is something that like the world has been seeing on the horizon for, I don't know, a decade, at least like very acutely two or three years. Um, and the missiles, in the missiles thing in particular, like we had, you know, four months to process, right? So I guess I am, um, I don't know, I, I spent a while not believing in this stuff, but like the, the capacity for, um, the American sort of like regulatory system to keep up with is something which is moving, um, even faster than what we've seen because basically like the regulation hasn't mattered, um, really until, until mythos. And then now all of a sudden it does. And like if the sort of circle of things that starts mattering for, um, continues to expand, um, you know, even doing the sort of one off stuff that we're going to see now, it's just like, I'm getting more pessimist to Chris that we're going to be able to, uh, we're going to be able to handle this and, you know, get through. There's technical questions here that you and I don't have the robust answers to, which is like how problematic was this particular incident? Um, I don't know. And, and that's important. Um, and, and regardless, even if it's not problematic, like, you know, the government and the company should work to resolve even minor things, but it does matter for like, okay, is this something that needed like, you know, intermediate, you know, fix within half a day or something where there could have been, you know, a little more time to discuss things. I don't know. And that also matters some for like, what's the future of this? Like, you know, how prevalent are, you know, critical jail breaks, um, which, which is kind of the key question here. Um, if, if you can get to them to a point when they're not super prevalent, um, then regulations seems pretty possible, right? Uh, but, you know, it has to be collaborative, but it's, but it's possible. Um, it does require the, all the companies that are making models of that level of capability to be both within the jurisdiction of the US government and, um, you know, having a, you know, good, collaborative relationship with the US government. Um, that is, you know, putting this issue aside, again, that is the case for, you know, largely the American companies. Um, it's certainly not the case for anyone developing models, you know, outside the US. It's not the case for Chinese companies. Um, I think there's also big, like, really big questions for what open source policy will be. I think it's pretty clear if you're just taking this incident at face value, like, I don't see how you could possibly permit an open source, open source mythos, um, under the, you know, way the administration has reproached the problem. So what does that mean? Does that mean like open source is just going to be told to stop in, you know, eight months or whatever, whenever it catches up to, to miss us, probably less this point. Um, but let's say they're a year behind. So around eight months, um, like maybe, right? Like that might be what happens. Um, and this is to a degree, but a problem that we've seen coming for a long time. So I don't know, like, I think if you actually have a big enough regime, right, where you actually can make sure that like, the US companies are pretty far ahead and, and, uh, you know, you're, you're, you're, you know, not, um, even though you're never going to be able to control like everyone outside, but if like they're delayed by a long time behind the US companies, and then you actually do have like good, um, you know, predictable measures and good relationships with the companies and the government to be able to like iron out things quickly whenever they, they, you know, put out things that are good and then iron out things quickly whenever they happen. Like that, that might actually be pretty good. And I think a lot of this is going to rely on AI actually getting better at fixing the problems. But like we're seeing that with RSI too, right? AI is getting better at making AI, like it's also going to get better at preventing jail breaks or things like that. So I feel like there's a, there is a path here to a world that is like, um, relatively safe and extremely prosperous, um, and very good for America, but it requires a, like a lot of smart regulatory steps on the domestic and international side to work. And I think we're seeing the pitfalls of what happens right now. If you don't have that, um, it's, it's pretty scary. Yeah. I mean, I guess I'm, uh, I'm almost thinking more broadly of like, okay, so what other sections of society are going to need their own project glass wing where like you get all the adult players to like really get on their ship six months, nine months, whatever before the stuff is released. And you know, for all of the hacks we see of corporate America, like I have a decent amount of faith in CISOs of, you know, Amazon and Google and Microsoft and, you know, financial institutions to like use that time wisely. Now, if we're sort of like going down the, um, you know, if we start looking at like different corners of, um, society, right, which aren't necessarily as well capitalized, which don't have as much technical talent where the sort of like earth shaking impacts of what a new model could bring over the next two years, um, that sort of, you know, that, that lead or that sort of breathing time where the quote unquote, like good guys have access to the models, but the, um, but sort of malicious actors be them, uh, states or criminals or whatever, just like can't, can't outrun whatever's on the open market from a closer open model perspective. It's just, I don't know, it's just, just, just frees me out a little bit Chris. Yeah. I, I, I, you're, I think you'll just have to have some normative changes, right? Like we're just going to have to be in a more security prioritized, you know, culture, right? Like you, there's just going to have to be an expectation that basically any product that ships is going to have to have gone through that, like consumer's use is going to have to go on through some degree of like AI, um, enabled like cyber defensive, you know, testing or upgrading. Like it just, it's just because otherwise is, you're just going to be smoked like right out of the door, right? So, um, but if you do that, and you have that kind of big lead, I actually think that in the cyber scent, we can actually get to a, a place where we, yeah, I mean, it's not just the cyber dynamic in cyber. You're right. It's so outside of cyber actually. I just think like if you're expanding the circle to the bio stuff and whatever, yeah, yeah, yeah, yeah, yeah, yeah. Because because like the cyber problem, I think we can kind of, you know, if we do it well, we can address it and then you can actually have like really, really good defensive systems and almost like getting advantage. But the problem is like you can't deploy AI based bio defenses in the way that you can, like as rapidly as you can with AI based cyber defenses. So because those are like physical things that you're going to have to get out there. So like that, that's where, uh, yeah, we're just going to have worth. They're just models that get just good, get good enough to like put entire towns out of work because they've like completely, you know, I know these CEOs, they keep talking about, we want, we don't, we want to create more jobs, not eliminate them. But, um, it just seems like there's, there's real prospects for like radical economic disruption. Even if it's not like the entire economy shuts down, like there are presumably corners of it where like 10 people can spin up companies which like do the job of, you know, 10, 100, 500,000 factories or, I mean, it's not just the AI labs that are saying this too. Like Jamie Dimon said, anyone who doesn't think the AI is going to cause job loss gets, I just got to get their head out of the sand, like it's, it's actually a lot of people who are following it. Obviously, I think that there's, there's obviously a segment of people who nest to maybe don't believe it or maybe want to push the narrative that, uh, you know, won't cause job loss to be up until the music stops so you can kind of keep going for as long as possible. Um, but, but it clearly will. The question is the magnitude. I don't think we totally know, um, and, and, and, and kind of when that will happen. But the idea that it won't at all is, is, is, is crazy. I mean, I think that's kind of why you're actually seeing this being a more bipartisan issue on Capitol Hill. Like I think actually like representatives are who are, you know, maybe not the most keenly aware on technology issues, but like when there's a prospect of massive job loss in their districts, all of them get very keenly aware, very quickly, right? And like, yeah, they're all paying attention to this, right? And actually that's why I think you're starting to see a regulation as a something that both parties recognize as necessary because if this actually happens, like, no one wants to be the guy who was, you know, supporting the, the, let the private sector rip policy that then, you know, replaced entire towns in their district, right? Yeah. So saline, which maybe is just like reverse political incentives, but regardless, like that is definitely how it's being viewed now. And I'm, uh, on the other side of getting the data center built in your town, um, saline Michigan, open AI very proudly announced was getting a $10 million refurbishment of their rec center. There you go. So, you know, peanut, some peanuts for the masses, it's a little help you, help you. This women classes will help you stay out of the underclass. I mean, but this is where the federal government should step in, right? Like this is where, like, yeah, we don't want to stop data centers, my opinion. Like, we, we, you know, that would cause, you know, all sorts of, you know, we lose our lead, blah, blah, blah, but like, we actually need a solution to help the people that are impacted by this. And the good news is the, the, the, the, the, yeah, the silver lining here is like, regardless on exactly what happens, this technology is going to create, like, very large amounts of wealth. The question is, who is it going to create it for? Is it going to be for like three to five people or is it going to be for the, everyone in the country? It's actually, the situation where it wasn't creating wealth, it would be, it would, we wouldn't even have an outcome, a, a path here, but it is, it's good. Yeah. Yeah. And sort of relying on corporate, large S and, um, you know, Dario, blog posts saying, yeah, we're really concerned about future post economic America and we're going to be, you know, trying to keep our janitors employed even after we turn them into robots is like, not necessarily. Like, no, this is like a job for legislators, not a job for CEOs to figure out. Yeah. Um, a one bit, you mentioned the, the, the Chinese models, uh, Jirupu put out a, uh, press report saying cutting edge intelligence should belong, uh, to not only a few, and it should not be withdrawn at any time. It should be open available, extensible and built to serve every developer, which like, you know, I don't doubt that the Jirupu AI CEO believes this, but, um, if I am a, like, I would also be shocked if, um, uh, when sort of like mythos level capabilities start coming out of these Chinese labs, if the Beijing, uh, feels similarly and not necessarily being cool with anyone and their mother get having access to them. Yeah. And I also think that we, we're having all these conversations about like chips in China, which I find frustrating because this is a debate that was kind of largely seldom 22 and we're really litigating it. But what that's also doing is it's like taking the oxygen out from some of the more interesting debates. I think people are talking, we're kind of realizing more now and need to be addressed. And I think one of them is model access. Like I actually do think that restrictions on model access are necessary and probably are necessary right now. I think that, you know, the reason that the RSI loop is closing is because the models are getting really good at R&D. So you probably don't want the Chinese AI labs able to use the USAI models to be able to do their R&D, which is what they're doing, right? Like our labs use their own models to ins, to expedite their R&D. Chinese labs use US models in order to expedite their R&D. So we probably should be talking about model access restrictions on China and on Chinese companies, you know, around the world and like, I don't know, have a, you know, we have to think about how to do that, but in a way that is like smart and targeted and isn't like just draconian, you know, resulting in the model going away, like that, that doesn't achieve the intended effect. But like, I don't know, I think the premise here is something that I think from a national security threat, we should be thinking about more, but we're only focused on American companies and not, you know, our adversarial companies. That's the problem. Can we talk a little bit about the, I don't know, yeah, if I'm a Canadian policy maker, European policy maker, Japanese policy maker, and I see this happening, I mean, they were already sweating about the sort of like drip, drip access that they got to mythos and then to sort of see this cut off. I mean, the irony is that like there isn't really another option for them, right? These companies are not going to like, I'm sorry, like no disrespect to like, you know, wimpy mistral over there, but I don't, I just don't think it's going to get it done. And so you're sort of either stuck just being a taker when it comes to American models or sort of just like waiting for China to deliver you something which, you know, as Chris and I both believe may end up starting to close down once they get powerful, more powerful enough. How do you see this sort of playing out in other capitals and like, to what extent does it actually matter? They definitely should be nervous. I mean, just because, you know, the fact that the administration imposed controls on all of our allies would inherently make them nervous. I do think that we are moving, even this particular situation aside, like we are probably moving towards a tiered system of model release just in general, whether and and I imagine that yeah, US, you know, tier one US companies will probably, the government and tier one US companies will probably be at the top of that list. And then you'll go down it from there to like, you know, tier two companies and then maybe close allies. Like, I kind of think that is how this is going to inevitably play out. I think that there's still a lot of, you know, that is basically how every other technology works, right? Like that's how all the most advanced weapons technologies are too, right? So that isn't necessarily unusual. I think it's, we've been very slow to build a regime around this. I think you started to see, you know, at the end of the Biden administration, there was some on efforts on, on kind of global infrastructure and, you know, putting some restrictions on chips on, you know, most countries other than say the 18, you know, the fact that that got pulled down is, it was very lucky for allies, but also like completely out of sorts with the Trump administration, right? Which generally speaking has been looking to, you know, exact leverage over allies and squeeze them. And so it's very strange that the one area where they decided to kind of give allies everything that they want was in AI, which is normally the most important area of future of technology and economy and military. But no, I think that we're inevitably, we're going to be moving towards a more restricted world in models and in infrastructure. The next question you asked though is, is how much does it matter? I mean, it certainly matters for them and, you know, where they take, you know, their technological development. I guess from a US China perspective, yeah. From what? I guess from a US China perspective, like how much does it matter if, you know, for both of them, I mean, I think, look, allies don't have a real recourse to this. And actually, it's still pretty good to be, you know, on the US, in the US ecosystem. Even if you don't have like the Tier 1A access, like it's still going to be far beneficial to being in the Chinese ecosystem. And I think we should make sure that, like obviously you want to make sure that everyone is, you know, doesn't, isn't incentivized to actually move to the Chinese ecosystem. But like they're all, you know, ultimately, like, I don't see that happening. I don't see allies working with China to build an alternate semiconductor supply chain and alternate AI model supply chain, all of which is going to be like, you know, full of Chinese propaganda and ultimately like used to undermine them. It's tough. I mean, this is, this is, you know, the US is like very dominant in this area. And I think it's hard for allies to accept that. But it's true and I think there's ways that that that can manifest in collaborative ways, even if it doesn't mean that just we're absolutely sharing everything freely. For most things, we don't share absolutely everything freely and still, it's far beneficial to be in our orbit than to be in Russia's orbit or China's orbit and it's gonna be the same way with AI, I think, over the longer term. - Yeah, I mean, like the most things, right? Like this is a giant commercial technology, as well as, you know, the frame Daria likes like nuclear power, nuclear bomb adjacent. So the fact that like we're ending up in some like awkward new middle, right? I mean, we weren't regulating electricity, wasn't a thing. But it seems like there's, you know, we're somewhere in between sort of like nuclear power and electricity and I don't know. Like in so far as this doesn't slow down, this doesn't get completely commoditized, then yeah, that like awkward, I don't even want to call it a middle ground. Yeah, that awkward sort of like I tar, but you know, way higher stakes and, you know, hopefully tuned in a way that like, you know, accounts for, you know, keeping, keeping the rest of the world happy and playing with American models and hyperscalers while not, you know, being too frustrated that they end up looking for another answer from the PRC. You know, these are all coming back to the beginning. Like, look, I want Jalen Brunson on the case with this, you know, not necessarily some, some bench players, it's gonna be, it's gonna be really hard to sort of balance all of that as you have the sand shifting under you of this, of these technologies developing in a way that, you know, you can't really predict even, you know, six or 12 months out. Yeah, I think the only thing you can really predict is that like, you are going to, having a big lead and a large lead is gonna be more important in a year than it is now. So like, this is only going up in capability and importance and in complexity. So the value of generally taking actions to, you know, improve security and improve your lead really compound over time. And we're paying the price for that now, right? We're paying the price for kind of inaction for the last, you know, year and a half largely. Because in now we have like very powerful capabilities, no regulations, chaotic environment, we're scrambling to catch up. And meanwhile, our adversaries are still kind of freely using our own technology. You know, even if we do cut that off, it's still gonna be, you know, some time until you start to see the large effects because there's a lot that's already out the door. But there's also no better time to start fixing all those things in now because all these problems are gonna be twice as hard four months from now. Yeah, the motivational Christmas choir here. Keep it, keep it, keep it the positive energy and times of dark policy making. So we had a, you know, the Friday afternoon and throughout the sort of BIS letter. I guess was it a week before? Oh my God, only that we had a Sunday letter from, from BIS trying to close a loop. Well, why don't you give us the context on all of that? I think it was two weeks before. Okay, yeah, it's two weeks before. Yeah, but no, this is just interesting. I mean, first of all, this is like a massive story. The fact that there has been an export control story that has leapfrogged this one in a period of 12 days is remarkable, but such are the times that we're in. No, so effectively, there is a emergency guidance that was issued two weeks ago by BIS to state that Chinese subsidiaries of Chinese companies located outside of China were not allowed to buy controlled AI chips. The reason that they had to issue this guidance was a loophole had been created by commerce that was seemingly permitting these orders. So Chinese, you know, China's been banned from receiving AI chips since 2022. Since 2023, that's been extended to Chinese companies operating anywhere in the world. So the obvious, you know, if you're going to ban them into 10 cents, then you don't want to allow 10 cents to buy them with their Malaysian subsidiary, right? Pretty logical here. But basically what happened without going into too much of the regulatory gory details, which I'm happy to, if you want to Jordan, but we'll try to spare people, is because the Biden administration in the AI diffusion rule, it replaced the explicit part of the regulation that captured Chinese headquartered companies operating overseas. It replaced that with a global license requirement. So it just, so it put forward a global license requirement that applied to anyone everywhere for AI chip orders. There were then some exceptions and exemptions for it. But as a matter of regulatory housekeeping, it struck the provision that captured chip exports to China headquartered companies, because just because it was redundant, the global control now captured that. So you didn't need two different parts of the regulation that would do that. The Trump administration said that they were not enforcing the global license requirement. They said they were going to replace the rule that was over a year ago, they still haven't. As a result, you know, talking with people in industry, I can attest to this, eventually companies started to say, okay, well, we're just going to take the law literally, because if you look at the regulations, there is not an explicit prohibition against us sending to Chinese headquartered companies overseas, which means that they could receive blackwell chips or even forthcoming Rubenships, they could have without a license. So that's a huge problem. And that would, I mean, it effectively neuters all of export controls. There's actually a second part of this, too, that meant not only that, but they could also make chips at TSMC or Samsung without a license as well. So we were in a very kind of bad situation. I don't know who exactly took advantage of this or in what magnitude that information, you know, is still yet to be revealed. I know a lot of people are looking for it, but what I do know is that companies were telling people that this was a legally permissible route to take. And there's a lot of rumors that there were some things that were shipped pursuant to this loophole. So commerce closed this two weeks ago and said, "Okay, no, you can't actually ship to Chinese headquartered companies." That also had the effect. Let's stay here. Well, what is this? The fact that it comes out on a Sunday tells you what? Well, is that this was a huge problem? Right, like, yeah, this isn't issued guidance a lot. The issue guidance on Sunday is almost never. Obviously, it was never the administration's intention to allow these shipments, right? The president's been very clear, like, no black wells. You know, we even, the president has like, this is like, tax system, right? Where the, the, say what you will about the policy to export ships to China. Obviously, I'm a big critic of it, but like, there was a system set up to-- - At least get our 25%, yeah. - Right, yeah, the system set up revenue for it. And like, this is a way to circumvent that. Like, this in no way is something that the administration intended. But the fact that it was happening shows that this is, this was a real problem that had to be once, you know, it was brought to administration's intention. I think they moved to close it right away. Generally speaking, I think this shows that the way that export controls have been administered has been, you know, incoherent. And has, it is not something that's really been prioritized. You know, the fact that you could create a loophole that would allow like this kind of fundamental undermining of the entire regime because you said you were not enforcing a rule that then had this cascading effect is crazy. I mean, actually, US export controls, if you look at them still, look at the regulation still for what type of chip is controlled. It will still say that there's a global license requirement for AI chips everywhere. And like, the AI diffusion rule is still on the books. Now, the administration says they're not enforcing that, but then if you're like, okay, well, what is the policy? What is the export control policy, AI chips? You would actually have to go higher a law firm to tell you exactly what is and what isn't because there's not, you can't even go back to like see a previous version of the regulations and like the way that the diffusion rule worked, it wasn't just, it didn't just modify, it wasn't just like stick one rule into the regulations. It modified things throughout the regs. So when you say we're not enforcing, like which provisions are you enforcing? Which one aren't you? No layperson could possibly understand that. So you have this crazy situation where like, it's impossible to tell what the export control regs are. And though it's not surprising that then companies find ways to work around them. So this is like, this is, we're girls to where you are on policy. It's patently unacceptable for the Department of Commerce regulations for over a year to not actually reflect the administration's policy. Like just decide what your policy is and make the regulations reflect that. They actually still reflect the Biden administration policies that the administration said that they wouldn't enforce. So this is what happens. So like you need to, you need to actually take this seriously. I think also this current episode shows that like export controls are super relevant and also need to be taken seriously. And the administration just hasn't been doing that. You're like, I don't have so much energy for it. No, it's, I mean, it's, it's just, I remember when they first said it, like we're not enforcing this rule. I assume that meant a week later we would get a new rule, right? Like that, and, and, and here we are. And we're kind of reaping what, what we sow I'm like, I guess just like not doing anything or not being able to agree on a thing or, you know, in the irony of all of this is like, they're so worried about doing something that would lead to some, you know, new escalation with China. The directive clearly from the top is like, don't do anything that would sort of like, change the balance that we reach as of October of last year. Fair, but that means that you don't go backwards for like no reason or like not as, like this wasn't like part of a grand bargain where like, okay, now they're going to, I mean, I don't know, like give us a trillion dollars of investment or something. Like it was just a sort of like bureaucratic fuck up that clearly they like clued onto on a Friday afternoon and then had to spend the weekend dealing with. I mean, the other question is like, how many chips had to get sent before they realized that this was a problem? How much, you know, you know, how much, how much tape out happened at TSMC for, for Huawei or whoever? I mean, it's just, yeah. I mean, there's some real issues over there, I think. They wouldn't know because there's no licenses required, right? Actually, no license required is even worse than approving licenses because all of the companies, like they didn't even have to tell the government how much they were shipping or what they were doing. It's a bad situation and also like these controls weren't even China controls, right? Like they were actually controls on other countries, on third countries and this is why it's so perplexing to me that like there's not more efforts to crack down on these because actually the China regime is largely in place but the problem is like Chinese smuggling is enabled through third countries. So the solution was mostly to like make sure that there's some, you know, license requirements on large numbers of third countries to limit that. You know, I think that's very clearly necessary now and that actually is not within the realm of the broader US-China, you know, Daytona, right? Like if we're saying, okay, we're gonna impose a license requirement on Malaysia or Singapore because obviously a preposterous numbers of GPUs are going there illegally to China. Like China can't really complain about that, right? Like they know that they're illegally smuggling the chips through these countries. Like we're just saying, okay, we're just, you know, preventing this obviously illicit behavior. That would actually diplomatically be achievable to do. I think the reason that these haven't been closed is because this has just been a whole issue set that's been deprioritized and ignored. But then as a result, actually the situation has degraded much farther than people think because we're actually allowing like, you know, completely open spigot sales of the most advanced chips to Chinese companies. And I should also note that actually there's one element that this hasn't been fixed. So in conjunction with the AI diffusion rule separately, there is this rule released called the Foundry Due Diligence Rule, which basically said for anyone who's sending an AI chip design to TSMC or any 14 nanometer chip design has to effectively confirm that it's not an AI chip through there's a few ways that they could do that or get a license. And the reasons like not that many companies are sending like really advanced designs to TSMC. So I think that rule has a white list of companies that are all permitted. And there's various other kind of routes you could go through but it's necessary to prevent like third country, third country, you know, cutouts from making chips at TSMC which we also know China did. So like Huawei got cut off but then you can get a cutout in, you know, Singapore or Malaysia or someone else to send a chip design to TSMC and then have them route through there. This is actually still permissible. So the Chinese company itself cannot send. That's now restricted since Sunday but the cutout can. So you get some random company in Malaysia, you know, not 10 cent ink can send a chip design to TSMC for a black well level chip and TSMC can, you know, unless they have, you know, very clear knowledge that's a Chinese company which does generally like, I don't know, no company is, is, you know, imposes like really strong, you know, enforcement measures on this. That's generally the lesson we learn not singling out, you know, any company here but like in theory, they could do that. So like the whole point of this was to cut down on that and that loophole was not closed by the guidance. So the administration is still not enforcing that rule and we'll see if they do that. But I really hope that they do, that 'cause like, I understand that a fusion rule was, you know, there's a bunch of, the administration wants to do their own thing, that's fine. But they never said that they were not enforcing the Foundry Due Intelligence Rule. They never said that they were not interested in enforcing that rule which is only exist to prevent Huawei from making chips at TSMC and Samsung and others. But the net effect of their actions has been not enforcing it and absent additional guidance. It will continue to not have a practical effect. - Let's close maybe on the kind of congressional response to the Trump administration. Can we even say high and cold relationship to kind of semiconductor export control policy? So what have you seen bubbling up from the House and Senate and what's interesting about it? - Yeah, I mean, there's a ton of movement on a lot of bills on this that are really granular too and really, you know, in the weeds, I think that there's, you know, on both sides, there's your bipartisan bills but the Republican-controlled House Foreign Affairs Committee has passed, you know, 18 export control bills out of markup in the last couple months. It was the kind of biggest swath of export control bills ever. There's a big push to get a bunch of them in the National Defense Authorization Act at the end of this year. So you're seeing a very heavy interest from Congress in this space. I think because there is a lot of concern on the China side, there's a lot of concern about AI and the nexus of them is proving that this is actually still a pretty bipartisan issue. The biggest ones are probably the Match Act and the AI Overwatch Act. The Match Act would be focused on equipment controls and would basically force the administration to impose equivalent controls on allies or negotiate or impose equivalent controls on allies as they have on us, which I think would cut off, you know, most DV, all DV shipments to China and a few other places where allies don't have the same level of control as we do, which is a big problem. The whole point is actually not to impose more controls on U.S. companies but to make sure that allied governments are doing the same as we are on control level, also on servicing and maintenance. There's all sorts of areas where, you know, allies have worked with us on this, but they're not imposing the same level of control as us. You know, Netherlands and Japan are still allowed to ship items to Smith South, whereas U.S. companies are completely out, but they still have some things they can do. They can also ship relatively advanced things like some DV machines to basically all fabs in China except for a handful. So there's a lot of advanced allied equipment going to China and there's also some allied equipment that's going to the most concerning facilities and this would basically stop all that. The other one is on AI Overwatch, which would look at GPUs, exports and basically ban blackwell exports to China. There's different versions in the House and Senate and you would kind of different provisions, but it's building off of the Gain AI Act also that almost passed in the NDA last year was in the Senate version, it was one of the last items being negotiated, but that would have said, basically a right of first refusal for U.S. customers over Chinese customers for AI chips, which in effect would have been a ban because the demand for them in the U.S. is far more than the supply. So and this would just be a straight up restriction, at least for the next few years on blackwell or more advanced chips. So like you're seeing a real push from Congress on this, I do think that some of one, at least there is going to be some export control bills in the NDA, I think we're so, this is getting so much attention and there's so many kind of like serious policy makers on both sides who are getting behind these things that you'll probably see something in. The question is like, is it some kind of token, you know, one of the smaller bills is kind of a token nod to it or is it one of the bigger bills? There's a few others that I haven't mentioned, the Remote Access Security Act, the Chip Security Act, which would have real impacts but lesser so than those two. And then there's a slew of smaller, you know, important but more minor ones as well. So that's a state of blame. - The one, the most interesting thing bit for me here is, you know, from a U.S.-China economic policy, Czech policy side, we're at a bit of a stalemate, right? Where the Trump administration is scared anything, they put on is going to sort of trigger rare earths and, you know, have real impacts for the U.S. ecosystem. But the dynamic is different when it comes from Congress. And of course, this is like a great American tradition of like hawks in Congress. I mean, you saw this during the Cold War with the Magnitsky Act and, you know, all this stuff for Jewish immigrants in the Carter era. You saw this more recently in sort of Obama era of sanctions escalation with respect to Iran. And having it come today, it changes the dynamic for the executive when they're sort of doing their, sort of bilateral negotiation with the, with whatever country it is, because then they get to say, look, you know, our hands are being forced here. We, this isn't, the NAA is not something we're going to go veto on behalf of China to keep you. you guys having access to various controls and it's also something you know the Biden administration was very anxious about getting the Japanese and Dutch and South Koreans upset but you know that changes when it's like the you know when it's when it's the House and Senate doing the squeezing. So yeah interesting to see to what extent that that dynamic ends up playing out where we see more of the kind of like China hawk stuff ultimately coming out of Congress for the remainder of the Trump administration as opposed to the White House. Yep that's certainly how it's playing out. I think that's right. I think it's why you see that you know I've heard that Chinese government is lobbying really heavily against the match act because they're concerned about it which is all more reason why exactly why I should go through but yeah no I I think that's right and the administration could kind of strategically use this right and you know they don't have to go out and actively you know proudly declare their support for match or overwatch or other things but they could just kind of quietly you know let it go forward and give a wink and a nod to people and then when it signs and say to China like oh sorry I heard it's a tie exactly like you said I think it's a totally viable approach also something that many administrations have done over the course of history that that that you know would allow them to create more negotiating space but but I don't know it's not clear if and how they'll do it I mean like look the big concern I still have is like the the approach that the administration is taking on export controls it seems like on AI generally they are moving at least in a direction that is directionally correct right they're moving more towards a this is a serious thing that needs to be taken seriously it's a big thing it's moving fast we have to have the adults running it we need to figure out how to get our hand around this it's really scary and hard and like obviously there's been a lot of missteps but like that is the general direction of travel also some good things like I'm not you know criticizing there that is the general direction of travel but the fact that the only export control related action they've taken on AI is this like extremely draconian one against American company says like it seems to be there there's some like institutional issues on on how export controls are being run and regulated at the same time as you have two weeks earlier this like unbelievable loophole that was created because of kind of export enforcement screw up on export controls there there just seems to be like a big disconnect between how it is thinking about in applying export controls and how it is thinking about in applying AI policy and that's really problematic because export controls are so so important to AI policy as I think the administration discovered over the weekend so like we need to fix this because it's not working right now yeah so maybe you're too nice to say it so I will um uh you know a lot Nick we've currently got him the prediction markets of like 50% to leave over the uh at the end of this year I remember at the beginning of the administration uh Kessler was a pick that a lot of people were excited about because it's like look it was this was in the context of like Matt Gaetz right um where oh here's this lawyer who's like worked adjacent to this stuff and has a totally normal functional resume that you would see being appointed in like a Romney administration or whatever um and now we've had a um I don't know it's like a uh the sort of the the ineffective policy making is one thing and that's something that is probably you know from a cabinet level on down but uh doing but but but the loophole is just like some serious serious um bureaucratic malpractice which is not necessarily something that you can chalk up to sort of to sort of she trump dynamics or managing the broader relationship it's just like a giant fuck up that human beings are responsible for not being on their ship yeah the questions how much was exploited like I don't know but like uh all you know Congress should subpoena all the OEMs and OEMs and figure out like who actually ship because it to be clear Nvidia and AMD probably actually didn't ship it almost certainly didn't ship through the loophole because for various bureaucratic quirks they were separately restricted because of a separate letter that was sent but like all the server makers who actually are the people who ship to the data centers you know super micro um you know Dell aces gigabyte like I'm not you know all those companies theoretically were unrestricted I don't know if any of them did or not but like yeah Congress should subpoena all of them and see who what their their logs were and who they shipped to and if any of them um shipped to directly to Chinese subsidiaries um we got to we got to we got to get to the bottom of this um but also just like this we just need kind of competent enforcement and controls I mean it's crazy like the the Bureau of Industry and Security has not issued a single technology based control on China since the start of the administration uh there's only one countrywide control on on China that they've put in effect for this duration as a administration and that was the affiliates rule which obviously got taken down uh you know in conjunction with the rarest control so there's been no countrywide measures on China um there there's there hasn't been also anything on on kind of the broader AI set at all um you know you keep in mind like the administration actually controlled model weights in AI diffusion uh partly because we saw like oh this was something that we're gonna have to think about you know the spread of these overseas the Trump administration is de facto controlling model weights like very clearly right like mythos cannot be exported overseas I mean even before this letter there was a tacit agreement about who could be who could be export to Scott Besson was like personally negotiating with Japanese banks which wants to get access like okay fine it's basically de facto export control like let's put an actual export control on the books so that people can see what it actually what is you know regulated and what's not everything has been done via these private letters but what that means is like most companies don't know that so okay those companies are restricted but like most companies who don't receive those letters are not restricted and also have no reason to know what the regulations are and this is how you get big screw ups like the uh like the the kind of subsidiaries loophole because it was actually completely valid for all these companies to ship to to Chinese and especially after a year where they said uh they were gonna replace the rule and then didn't so um yeah we just like I'll say this for the empty time but like really need to take this very important issue set seriously um you know maybe this weekend is kind of the the opening move on export controls then now we're doing a export controls now okay we're back like wish it was on Chinese companies and not on American companies but like you know okay well let's miss I don't know it's a there's one element of this that's good like but now let's like vector it in the right direction actually uses very powerful tool to support uh you know us companies and uh heard our adversaries all right um yeah I'll send it there Chris congratulations again thanks so much for your part time congratulations to you too we both deserve correct congratulations I think but yeah DC never read a page most powerful model in the world life Thursday dark by Friday Darry your car to the new civilization on the line beggin to regulate then I should get the sign high wolf for four years to the cuffs fit better jassy drop the dime in the admin grab the letter 521 they can't it killed the best in the game I'm 29 and folded whole bitch duck in the blame they never read the plate just reacted on the fly Nick study the tape and clothes that's the why close it chips roll everything god the GPU they never read the rule put the next red Joe the Nick one side cut the next down one side cut a deal New York Newtian Simon DC can't find the well can't find it run and read the set for the set even called Kessler signed a rule that's the dead pod case by case off the dribble squintin that is on paper weight what this do eight two hundred guards see you later downtown set the screen on time knew just where he's needed down the heart of leaky chips I never even read it ever ain't a no be in the lab when the sun ain't even up bitch wake up at noon loop for the size of Taiwan everything god the GPU they never read the rule put the next red Joe one side cut the next down one side cut a deal New York Newtian Simon DC can't find the well yeah same read the brief called the rules for phones now he's drafting the regulation scrambling for the tool for it strangled the future now it's his Lord of signs same rule he called a chokehold now it's brilliant now it's fine best it won't from the nap when the trophy start to clean wake up nope jensen slid it smooth let the jacket saw pitch talk trunk they're out the house but brunson don't bite no no chips rule everything god the GPU they never read the rule put the next red Joe one side cut the next down one side cut a deal New York Newtian Simon DC can't find the will can't find it the next watch film Newtian Simon they're still asking what the assignment was c-r-e-a-m-m-m diversify your export controls