The Day AI Got Banned: Anthropic's Emergency Shutdown

Alex Shannon: There are two futures we could be heading toward right now. In one, governments step in when AI systems get too powerful, companies comply, we fix the problems, and everything works out. In the other, we’re looking at a world where the most advanced AI models get weaponized, where safety measures become meaningless, and where the gap between what AI can do and what we can control becomes impossible to bridge.

Sam Hinton: And today, we found out which path we’re actually on. Because for the first time in history, the US government just forced a major AI company to shut down its most advanced models. Not slow them down, not add more safety measures - shut them down completely.

Alex Shannon: We’re talking about Anthropic’s Claude Fable 5 and Mythos 5 - models that millions of people were using until yesterday. And the reason they’re offline? Someone figured out how to completely bypass their safety systems.

Sam Hinton: This isn’t a theoretical risk anymore. This is the moment when AI safety stopped being a future concern and became an emergency happening right now.

Alex Shannon: You’re listening to Build By AI, I’m Alex Shannon, and we are diving deep into what might be the most significant AI story of 2026 so far.

Sam Hinton: And I’m Sam Hinton. Look, we cover AI developments every day, but what happened with Anthropic yesterday changes everything. We’re talking about the first major government intervention in AI deployment, international access restrictions, and security vulnerabilities that were serious enough to shut down billion-dollar systems overnight.

Alex Shannon: Plus we’ve got Mistral potentially raising money at a twenty billion euro valuation, and a Chinese cybercrime operation that used AI to scam hundreds of thousands of people. It’s a lot to unpack.

Sam Hinton: But first, let’s talk about what exactly happened with Anthropic, because this story has layers that go way beyond just one company’s problems.

Anthropic Says It’s Taking Claude Fable 5 Offline to Comply With US Government Order

Alex Shannon: Alright, so here’s what we know. Yesterday, Anthropic published a blog post announcing they’re taking Claude Fable 5 offline completely. Not pausing it, not adding more safety measures - offline. And they’re doing this because the US government ordered them to. The reason? Someone discovered a jailbreaking method that could completely bypass the model’s safety measures.

Sam Hinton: And this isn’t just Fable 5. We’re also seeing Mythos 5 get pulled, and the government has barred foreigners from using any of Anthropic’s most advanced models. This is unprecedented, Alex. I mean, we’ve seen content policies change, we’ve seen voluntary safety pauses, but a government order to shut down deployed AI systems? That’s never happened before.

Alex Shannon: What’s really striking to me is the speed of this. Companies don’t usually shut down their flagship products overnight unless something is seriously wrong. What kind of jailbreaking vulnerability would be bad enough to trigger this kind of response?

Sam Hinton: Right, and think about what jailbreaking means in this context. These safety measures aren’t just about preventing offensive content - they’re about preventing AI systems from helping with dangerous activities, from providing instructions for harmful things, from being manipulated into doing tasks they weren’t designed for. If someone found a reliable way to bypass all of that…

Alex Shannon: OK but here’s what I’m struggling with - Anthropic has always positioned themselves as the safety-first AI company. They’ve built their entire brand around constitutional AI and responsible deployment. If their systems have this kind of vulnerability, what does that say about the industry as a whole?

Sam Hinton: Dude, that’s exactly why this is so scary. If Anthropic, with all their focus on safety research and responsible AI, can have a vulnerability serious enough to trigger a government shutdown, then every other AI company should be freaking out right now. Because if it can happen to Claude, it can happen to GPT, to Gemini, to any of these systems.

Alex Shannon: And I keep thinking about the timing here. This jailbreaking method was discovered - by whom? Was this internal red team testing that found the vulnerability? Was it external researchers? Or worse, was it bad actors who were already exploiting it?

Sam Hinton: That’s a crucial detail we don’t have yet. If this was found by Anthropic’s own safety team or academic researchers, then maybe we caught it before real harm was done. But if this vulnerability was being exploited in the wild, if there were already people using it to get these models to do dangerous things…

Alex Shannon: Then the government response makes a lot more sense. You can’t just issue a patch and hope for the best if people are actively exploiting a security vulnerability in real time. You have to shut it down immediately.

Sam Hinton: And think about what this means for anyone who was building products or services on top of these models. Overnight, their entire business model just disappeared. No warning, no migration period, just gone. That’s a level of platform risk that I don’t think most developers were accounting for.

Alex Shannon: Right, because we’re used to thinking about platform risk in terms of API changes or pricing adjustments. But government-ordered shutdowns? That’s a whole different category of risk that changes how you evaluate the stability of building on any AI platform.

Sam Hinton: And the international restriction piece adds another layer. Barring foreigners from using these models suggests this isn’t just about general safety - this is about national security implications. The government is treating advanced AI models like controlled technology.

Alex Shannon: Which, honestly, maybe they should be. We’ve been talking for years about AI as a strategic technology, about the geopolitical implications of advanced AI systems. Well, this is what it looks like when that theoretical concern becomes policy. The US is essentially saying that these AI models are too powerful and too risky to let other countries have access to them.

Sam Hinton: But here’s what I find interesting - Anthropic is complying immediately. There’s no legal challenge, no pushback, no ‘we’ll fight this in court.’ They just said ‘okay’ and shut everything down. That suggests either the legal authority here is crystal clear, or the vulnerability was so serious that even Anthropic agrees this is the right move.

Alex Shannon: Or both. I mean, if you’re Anthropic and you’ve built your entire reputation on being the responsible AI company, and then your flagship model turns out to have a serious safety vulnerability, your incentives are actually aligned with shutting it down quickly and fixing it properly.

Sam Hinton: So what does this mean practically? For developers who were building on these models, for businesses that integrated them into their workflows, for researchers who were using them?

Alex Shannon: Short term, it’s chaos. Any application or service that was built on top of Fable 5 or Mythos 5 is now broken. But longer term, this sets a precedent. It tells every AI company that the government will step in if they think your model poses a risk. That’s going to change how companies approach AI development, how they think about safety testing, how they roll out new capabilities.

Sam Hinton: And I think this changes the competitive dynamics too. If you’re OpenAI or Google, are you looking at this thinking ‘thank god it wasn’t us,’ or are you scrambling to test your own models for similar vulnerabilities? Because if this is an architectural issue that affects multiple systems…

Alex Shannon: Then we might see more shutdowns. And that’s terrifying from a business continuity perspective. Imagine if GPT-4 or Gemini got shut down tomorrow because of a similar vulnerability. The entire AI application ecosystem would collapse overnight.

Sam Hinton: Which might be what needs to happen if these systems really are vulnerable in fundamental ways. I mean, we don’t let unsafe cars stay on the road just because it would be economically disruptive to recall them.

Alex Shannon: True, but cars don’t power millions of applications and services that people depend on every day. The economic disruption from widespread AI shutdowns would be unlike anything we’ve seen in the tech industry.

Sam Hinton: And keep an eye on how other countries respond to this. If the US is restricting access to American AI models based on nationality, you can bet China, the EU, and other major players are going to accelerate their own AI development programs. This could be the moment that really fragments the global AI landscape.

Alex Shannon: Yeah, and that fragmentation might happen faster than anyone expected. If you’re a government official in Europe or Asia watching this unfold, your takeaway isn’t ‘AI is dangerous’ - it’s ‘we need our own AI that can’t be shut down by the Americans.‘

U.S. Bars Foreigners From Using Anthropic’s Most Advanced A.I. Models - The New York Times

Alex Shannon: Let’s dig deeper into this international access restriction, because I think this might be even more significant than the safety vulnerability itself. The US government has essentially weaponized AI access, turning advanced language models into tools of foreign policy. If you’re not an American, you can’t use the most powerful AI systems developed by American companies.

Sam Hinton: Yeah, and this is happening while every other major economy is pouring billions into their own AI programs. The EU has been working on their AI sovereignty initiatives, China has been developing their own large language models, and now the US is basically saying ‘our AI is for Americans only.’ This is going to accelerate the fragmentation of the global internet we’ve been worried about.

Alex Shannon: But how do you even enforce something like this technically? I mean, VPNs exist, people can mask their location. Is this more about creating a legal framework than actually preventing access?

Sam Hinton: It’s probably both. Look, determined users will find ways around geographic restrictions, but this isn’t really about stopping individual users. This is about preventing foreign governments, foreign companies, foreign researchers from integrating these models into their infrastructure, their products, their strategic systems. You can’t build a VPN-based business model around someone else’s AI if access could be cut off at any moment.

Alex Shannon: Right, and think about the implications for multinational companies. If you’re a European company that’s been using Claude Fable 5 for customer service or content generation, you can’t just switch to having your American subsidiary access it for you. That creates legal and operational complications that most companies aren’t going to want to deal with.

Sam Hinton: And there’s the broader signal this sends about how the US views AI in the context of international competition. We’re essentially treating advanced AI models the way we treat advanced semiconductors or cryptographic technology - as strategic assets that need to be controlled.

Alex Shannon: Which is probably the right approach, honestly. I mean, if these AI models are as powerful as we think they are, if they’re going to be central to economic competitiveness, military capability, social influence - then yeah, maybe they should be treated as controlled technology. The question is whether this kind of restriction helps American competitiveness or just pushes other countries to develop alternatives faster.

Sam Hinton: That’s the big risk, right? If you’re the European Union or if you’re China, and you suddenly can’t access American AI models, your response isn’t to give up on AI - it’s to accelerate your own programs. And maybe those programs don’t have the same safety focus that American companies have been developing.

Alex Shannon: Exactly. So we might end up in a world where the safest, most carefully developed AI models are restricted to Americans, while the rest of the world ends up using less safe alternatives developed by countries that prioritize speed and capability over safety research. That’s not great for global AI safety.

Sam Hinton: But here’s what I find fascinating about the enforcement question - this isn’t just about IP addresses or location tracking. Advanced AI models require significant computational resources, ongoing support, enterprise integrations. It’s actually pretty hard to hide large-scale commercial use of these systems.

Alex Shannon: That’s a good point. Individual users might be able to get around geographic restrictions, but if you’re trying to build a business or integrate these models into critical infrastructure, you need legitimate, sustainable access. You can’t run a company on VPN tricks.

Sam Hinton: And for American AI companies, this creates a weird dynamic. On one hand, you have government protection from foreign competition. On the other hand, you’re losing access to global markets, global talent, global feedback that helps make your models better.

Alex Shannon: Yeah, and some of the best AI research talent is international. If you’re restricting access to your models based on nationality, you’re also potentially limiting your ability to attract international researchers and developers who want to work with the most advanced systems.

Sam Hinton: Right, and if you’re an AI researcher or entrepreneur outside the US, this is a huge wake-up call. You can’t build your career, your company, your country’s AI strategy around American AI models if access can be revoked for geopolitical reasons. That’s going to drive a lot of investment and talent toward non-American alternatives.

Alex Shannon: And this probably changes how international tech partnerships work. If you’re a European company looking to partner with an American AI startup, you now have to factor in the risk that the US government might cut off your access to the underlying technology for political reasons.

Sam Hinton: So we might be looking at the beginning of the end of the global AI ecosystem. Instead of one interconnected world of AI development, we’re heading toward regional AI blocs with limited interaction between them.

Alex Shannon: Which might be inevitable with technologies this powerful, but it’s still a massive shift from how we’ve thought about the internet and global technology development for the past few decades. We’re moving from a world of global platforms to a world of national or regional technology sovereignty.

Sam Hinton: And the speed of this change is remarkable. I mean, just last year we were talking about AI as this global, collaborative research effort. Now we’re looking at AI nationalism becoming a defining feature of international relations.

Anthropic’s Claude Fable 5 and Mythos 5 AI suspended over security fears - BBC

Alex Shannon: Now, the BBC is reporting that both Fable 5 and Mythos 5 have been suspended, and they’re framing this specifically around security fears. What’s interesting is that we’re getting more clarity that this affects multiple model families, not just one system. Mythos 5 was Anthropic’s multimodal model - it could handle text, images, code, all sorts of inputs. So we’re talking about a broad vulnerability that affects different types of AI systems.

Sam Hinton: And that suggests this isn’t just about one specific jailbreaking technique. If the vulnerability affects both a text model like Fable 5 and a multimodal model like Mythos 5, we’re probably looking at something fundamental about how these systems process and respond to inputs. Maybe something in the underlying architecture that makes them all vulnerable to the same type of attack.

Alex Shannon: What worries me is that if this is an architectural issue, it might not be limited to Anthropic. Most of these large language models are built on similar transformer architectures, trained using similar methods. If there’s a fundamental vulnerability in how these systems work…

Sam Hinton: Then every major AI company should be scrambling to test their models right now. OpenAI, Google, Meta, everyone who’s deployed large language models should be asking whether their systems have the same vulnerability. And the fact that we haven’t seen other companies proactively shut down their models suggests either they’ve tested and found they’re safe, or they haven’t found the problem yet.

Alex Shannon: Or they found the problem and they’re trying to fix it quietly rather than shutting down their systems. Which, from a business perspective, I understand, but from a safety perspective, that’s terrifying.

Sam Hinton: Yeah, because look at the incentives here. Anthropic shut down their models and are facing huge business disruption, angry customers, competitive disadvantage. If their competitors can quietly patch the same vulnerability without shutting down, then Anthropic basically took a bullet for the whole industry’s safety problems.

Alex Shannon: But maybe that’s why the government stepped in with an order rather than just a recommendation. If companies have incentives to keep dangerous systems running while they try to fix them, then maybe external pressure is necessary to force the right behavior.

Sam Hinton: And think about what ‘security fears’ actually means in this context. We’re not talking about data breaches or privacy violations - we’re talking about AI systems that could potentially be manipulated to cause real-world harm. That’s a fundamentally different category of security risk.

Alex Shannon: Right, and the fact that both models got suspended simultaneously suggests this vulnerability was serious enough that there was no safe way to keep either system running while they worked on a fix. That’s not a minor bug - that’s a fundamental flaw in how these systems handle safety constraints.

Sam Hinton: Which brings up a bigger question about AI governance. Should we have a regulatory framework that requires companies to shut down AI systems when serious vulnerabilities are discovered? Should there be mandatory disclosure of jailbreaking methods? Should the government have the power to order these kinds of shutdowns?

Alex Shannon: I think yesterday answered that question. The government does have that power, and they’re willing to use it. The question now is whether this becomes a regular part of AI oversight or whether this was a one-off response to an especially serious vulnerability.

Sam Hinton: And whether other governments follow suit. If the US can order AI shutdowns for security reasons, what stops China from doing the same thing? Or the EU? We could end up in a world where AI companies have to navigate shutdown orders from multiple governments with different safety standards and political priorities.

Alex Shannon: Imagine being an AI company trying to serve global markets when any government in any jurisdiction where you operate could order you to shut down your models. That’s going to make international expansion incredibly risky for AI companies.

Sam Hinton: And it changes how investors think about AI companies too. Platform risk just got a lot more complicated when governments can shut down your core product overnight. That has to factor into valuations, into business planning, into everything.

Alex Shannon: The bottom line is that this changes the risk calculation for anyone building on or investing in AI systems. Political and regulatory risk is now a major factor in AI deployment, not just technical and business risk.

Sam Hinton: And maybe that’s appropriate for technologies this powerful. I mean, we don’t let pharmaceutical companies deploy drugs without extensive safety testing and regulatory approval. Maybe AI models that can influence millions of people need similar oversight.

Anthropic Halts Access to Top AI Models After U.S. Ban on Foreign Use - WSJ

Alex Shannon: The Wall Street Journal is adding another angle to this story - they’re reporting that Anthropic has halted access to their top models, period. Not just for foreign users, but completely. And they’re connecting this directly to the US ban on foreign use. So it sounds like the government basically said ‘if you can’t restrict this safely to just American users, then shut it down entirely.’

Sam Hinton: That’s actually a really important distinction, because it suggests the technical challenge of implementing geographic restrictions might have been part of the problem. Maybe Anthropic couldn’t figure out how to reliably block foreign access while keeping the models available for American users, so the safest option was to shut everything down.

Alex Shannon: Which makes sense when you think about how these AI models are actually deployed. They’re not like websites where you can just check IP addresses. These models get embedded into applications, accessed through APIs, integrated into workflows that might span multiple countries. Figuring out the nationality of every end user is probably incredibly complex.

Sam Hinton: And if the vulnerability is serious enough that the government is worried about foreign access, then you can’t take any chances with your geographic restrictions. You can’t have a system where someone in China or Russia could potentially access these models by routing through an American server or using an American partner company.

Alex Shannon: Right, and this probably reveals something about how integrated these AI systems have become in global business operations. If it was easy to separate American users from foreign users, Anthropic probably would have done that instead of shutting down completely. The fact that they chose a complete shutdown suggests the user base was too interconnected to cleanly separate.

Sam Hinton: But this creates a really weird precedent for how AI governance works. The government can essentially force American AI companies to choose between serving global markets and serving American users. If you can’t reliably restrict access to Americans only, then you have to shut down globally.

Alex Shannon: And that’s going to make American AI companies less competitive internationally. If your business model depends on being able to serve global customers, but the government might force you to shut down rather than serve foreign users, then international customers are going to be hesitant to build on your platform.

Sam Hinton: It’s like the reverse of the TikTok situation. Instead of foreign companies being restricted from serving American users, now American companies are being restricted from serving foreign users. And the long-term effect is probably the same - a more fragmented, less interconnected global technology ecosystem.

Alex Shannon: And think about what this means for the global competitiveness of American AI companies. If Chinese or European AI companies don’t have these restrictions, they can serve global markets while American companies are limited to domestic users. That’s a huge competitive disadvantage.

Sam Hinton: But maybe that’s inevitable with technologies this powerful. I mean, countries don’t let their most advanced military technologies get exported freely. If AI models are really as strategically important as we think they are, then maybe treating them like controlled technology is the right approach, even if it hurts business.

Alex Shannon: The question is whether this kind of restriction makes the world safer or more dangerous. If American AI models are the safest and most carefully developed, then restricting access to them might push other countries toward less safe alternatives.

Sam Hinton: Or it might incentivize other countries to develop their own safety research and governance frameworks. Maybe the long-term effect is that we get multiple independent approaches to AI safety instead of everyone depending on American companies to figure it out.

Alex Shannon: But there’s also the question of whether Anthropic had any choice here. The Wall Street Journal is framing this as a response to the government ban, which suggests this wasn’t entirely Anthropic’s decision. They were essentially forced into this position by government policy.

Sam Hinton: Which raises questions about the relationship between AI companies and the government. How much influence does the government have over these companies’ technical decisions? What happens if a company disagrees with a government assessment of risk?

Alex Shannon: And we don’t know how much warning Anthropic had about this policy change. Did they have time to prepare alternative solutions, or was this a sudden directive that forced them to shut down immediately?

Sam Hinton: Either way, this is a massive shift in how we think about AI as a global technology. The era of freely accessible, globally deployed AI models might be ending before it really got started.

Alex Shannon: And it sets a precedent that other governments are definitely paying attention to. If the US can essentially nationalize access to AI technology, what stops other countries from doing the same thing with their strategic technologies?

Mistral is rumored to be raising €3B at €20B valuation

Alex Shannon: Alright, shifting gears to some funding news. Early reports suggest that Mistral, the French AI company, is raising three billion euros at a twenty billion euro valuation. That’s nearly double their previous valuation from their Series C round.

Sam Hinton: If confirmed, this is huge for European AI independence. Here we are talking about American companies getting restricted by the US government, and meanwhile Mistral is raising enough money to potentially compete at the same level as OpenAI and Anthropic. The timing couldn’t be more perfect.

Alex Shannon: Twenty billion euros for a company that’s what, two years old? That puts them in the same valuation range as some of the biggest tech companies in Europe. And it suggests investors are really betting that the AI market is going to fragment along national or regional lines.

Sam Hinton: Exactly. If you’re a European business or government, and you’re watching American AI models get shut down or restricted based on US government orders, then investing in a European alternative starts to look really smart. Mistral might be the biggest beneficiary of American AI protectionism.

Alex Shannon: And three billion euros is serious money - that’s enough to compete on talent acquisition, computational resources, research infrastructure. Mistral isn’t just positioning themselves as a regional alternative anymore; they’re trying to be a global AI leader.

Sam Hinton: What’s interesting is that this valuation suggests investors think there’s going to be sustained demand for non-American AI models. This isn’t just about hedging against geopolitical risk - this is a bet that the market for AI is big enough to support multiple major players with different national allegiances.

Alex Shannon: Right, and if the US continues down this path of restricting access to American AI models, then Mistral and other non-American AI companies could find themselves serving not just their home markets, but all the international customers that American companies can’t serve anymore.

Sam Hinton: The question is whether Mistral can actually deliver AI capabilities that compete with the most advanced American models. Having twenty billion euros in funding is great, but you still need the technical expertise, the research breakthroughs, the computational infrastructure to build world-class AI systems.

Chinese cybercrime operation that used AI to scam ‘hundreds of thousands of victims’ sued by Google

Alex Shannon: Speaking of international AI concerns, Google has sued a Chinese cybercrime operation called ‘Outsider Enterprise’ that used AI to scam hundreds of thousands of victims. According to early reports, this group sent two and a half million text messages over just two weeks as part of their scamming operation.

Sam Hinton: Two and a half million messages in two weeks - that’s the scale that AI enables for cybercrime. You couldn’t run that kind of operation manually, but with AI generating personalized scam messages, translating between languages, adapting to responses, suddenly you can target hundreds of thousands of people simultaneously.

Alex Shannon: And this is probably just the beginning. As AI models become more sophisticated and accessible, the barrier to entry for these kinds of large-scale scam operations gets lower and lower. You don’t need a big team anymore, just good AI tools.

Sam Hinton: Which ties back to our main story, right? This is exactly the kind of thing that governments are worried about when they talk about AI security risks. It’s not just theoretical jailbreaking - it’s AI being used right now to commit crimes at unprecedented scale.

Alex Shannon: And the fact that Google is suing rather than just blocking these accounts suggests this operation was sophisticated enough to evade normal detection methods. They probably used AI not just for message generation, but for evading security systems, creating convincing fake identities, managing complex multi-stage scams.

Sam Hinton: Right, and ‘hundreds of thousands of victims’ means this wasn’t just about financial loss - this probably caused real psychological harm, damaged trust in legitimate communications, maybe even affected people’s willingness to engage with digital services at all.

Alex Shannon: What’s scary is that this operation was apparently sophisticated enough to require legal action from Google, one of the most technically advanced companies in the world. If Google can’t just technically solve this problem, what hope do smaller companies or individual users have?

Sam Hinton: And this is happening while we’re talking about restricting access to the most advanced AI models. The question is whether these restrictions actually prevent bad actors from accessing AI capabilities, or whether they just push them toward less regulated alternatives that might be even harder to track and control.

Statement on US government directive to suspend access to Fable 5 and Mythos 5

Alex Shannon: We’re also seeing other companies releasing statements about the US government directive to suspend Fable 5 and Mythos 5 access. While we don’t have the full content of these statements yet, it suggests this is becoming an industry-wide conversation, not just an Anthropic problem.

Sam Hinton: Yeah, and that makes sense. If there’s a fundamental vulnerability that affects multiple AI architectures, then every company in the space needs to figure out whether their models have the same problem. These statements are probably companies trying to reassure their users that they’re taking the security concerns seriously.

Alex Shannon: It’ll be interesting to see whether any other companies proactively shut down their models for testing, or whether they wait to see if the government orders them to do the same thing that happened to Anthropic.

Sam Hinton: The fact that we’re seeing coordinated industry responses suggests this vulnerability might be more widespread than initially thought. Companies don’t usually release joint statements unless they’re all facing the same problem.

Alex Shannon: And the timing of these statements is interesting too - are companies getting ahead of potential government action, or are they responding to pressure from customers and partners who are worried about similar vulnerabilities in their systems?

Sam Hinton: Either way, this is becoming a defining moment for the AI industry’s approach to safety and governance. How companies respond to this crisis is going to shape public and regulatory perception of AI safety for years to come.

Alex Shannon: Right, and these statements are probably being watched closely by regulators, investors, and international partners. This is as much about managing stakeholder confidence as it is about addressing technical security concerns.

Our response to the US ban on Fable 5 and Mythos 5

Alex Shannon: And we’re seeing responses to the US ban from various industry players, though again, we don’t have the full content of these responses yet. But the fact that multiple organizations feel the need to respond publicly suggests this is being seen as a watershed moment for the industry.

Sam Hinton: Absolutely. This is the kind of event that forces everyone to pick a side. Are you going to support government intervention in AI deployment when there are security risks? Are you going to criticize it as government overreach? Are you going to use this as an opportunity to promote your own models as safer alternatives?

Alex Shannon: And the responses probably depend a lot on where these companies are based, what their relationship with the US government is, and whether they think their own models might have similar vulnerabilities. This is as much about positioning for the future regulatory environment as it is about responding to today’s news.

Sam Hinton: The international dimension is particularly interesting. Are we going to see European or Asian AI companies criticizing American AI governance as protectionist? Or are they going to focus on promoting their own models as more reliable alternatives that won’t be subject to sudden government shutdowns?

Alex Shannon: And how do American companies respond to international criticism while also maintaining their relationships with US regulators? This creates a really complex diplomatic challenge for companies that want to serve global markets.

Sam Hinton: Right, and these responses are going to be studied closely by policymakers who are trying to figure out how to regulate AI without destroying innovation or international competitiveness. This is essentially a real-time experiment in AI governance.

Alex Shannon: What I’m curious about is whether we’ll see any companies defend Anthropic’s decision or criticize the government action. That would tell us a lot about how the industry views the balance between safety and business continuity.

BIGGER PICTURE

Alex Shannon: If you zoom out and look at everything we covered today, there’s a really clear pattern emerging. We’re moving from a world where AI development was primarily driven by companies competing in global markets, to a world where AI is becoming a tool of national policy and international competition.

Sam Hinton: Right, and the Anthropic shutdown is just the most dramatic example of that shift. But you see it in Mistral’s massive funding round - investors betting on European AI independence. You see it in AI-powered cybercrime operations - countries using AI as a tool for asymmetric conflict. You see it in the government’s willingness to shut down American AI models rather than let foreign users access them.

Alex Shannon: And I think we’re going to look back at today as the day when AI stopped being a technology industry story and became a national security story. The question isn’t whether AI models are powerful enough to matter geopolitically - it’s how countries are going to manage that power.

Sam Hinton: Which means everyone working in AI - developers, researchers, entrepreneurs, investors - needs to start thinking about political risk as a core part of their strategy. The technical challenges of building AI are hard enough, but now you also have to navigate an environment where governments might shut down your models, restrict your access to markets, or force you to choose between different national regulatory frameworks.

Alex Shannon: And what’s really striking is how quickly this change happened. I mean, six months ago, the biggest AI policy debates were about content moderation and bias. Now we’re talking about emergency government shutdowns, international access restrictions, and AI models as strategic national assets.

Sam Hinton: That acceleration is partly because the technology got more powerful faster than anyone expected, but it’s also because geopolitical tensions made AI a proxy for broader international competition. This isn’t just about technology policy anymore - it’s about economic warfare, military advantage, cultural influence.

Alex Shannon: And for regular people using AI tools, this probably means less choice, higher prices, and more fragmentation in what AI capabilities are available where you live. The global, open internet approach to AI development is ending before it really got started.

Sam Hinton: But maybe that’s necessary. Maybe technologies this powerful need this kind of oversight and control. The question is whether we end up with thoughtful, safety-focused governance, or whether we end up with a fragmented mess where every country has different AI policies and nobody can build anything that works across borders.

Alex Shannon: And there’s this weird paradox where the countries that are most concerned about AI safety might end up with the safest AI systems, while countries that prioritize speed and capability end up with more dangerous alternatives. So global AI safety could actually get worse as a result of these restrictions.

Sam Hinton: Right, and that creates pressure for international coordination on AI governance, but at the same time, these restrictions make that coordination less likely because countries are treating AI as a competitive advantage rather than a shared challenge.

Alex Shannon: The other thing that strikes me is how this changes the investment landscape for AI companies. Suddenly, geographic location matters a lot more than it used to. Being an American AI company gives you access to American markets but might restrict your international opportunities. Being a European AI company gives you access to EU markets and potentially serves customers that American companies can’t reach.

Sam Hinton: And talent flows are going to change too. If you’re a brilliant AI researcher, do you want to work for a company that could have its models shut down by government order? Do you want to work on systems that can only be used by people from certain countries? That might drive more international talent toward non-American AI companies.

Alex Shannon: But it might also drive more investment into American AI companies if investors think government support gives them a competitive advantage in domestic markets. This could accelerate the development of American AI capabilities even as it limits their global reach.

Sam Hinton: And we haven’t even talked about what this means for open source AI development. If governments are willing to shut down commercial AI models for security reasons, how long before they start regulating open source AI projects? Can you control the safety of AI systems if anyone can download and modify the code?

Alex Shannon: That’s maybe the biggest question of all. Because right now, these restrictions only apply to commercial AI services. But the underlying technology is increasingly available as open source models that anyone can run. How do you enforce geographic restrictions or safety controls on software that people can download and run on their own computers?

Sam Hinton: You probably can’t, which means we might be heading toward a world where commercial AI services are heavily regulated and geographically restricted, but open source AI development continues in a kind of regulatory gray area. That creates all sorts of weird incentives and competitive dynamics.

OUTRO

Alex Shannon: This has been Build By AI. I’m Alex Shannon, and honestly, today felt like one of those days where we’re watching history happen in real time. The AI industry just changed in a fundamental way.

Sam Hinton: I’m Sam Hinton, and yeah, this is the kind of episode we’ll probably reference for months as the moment when AI governance got real. If you found this valuable, make sure you’re subscribed because we’re going to be following this story closely as it develops.

Alex Shannon: Thanks for listening, and we’ll see you tomorrow with whatever comes next. Because at this point, who knows what’s going to happen next in AI.