What the Anthropic shutdown means for Europe

0:01Hello, this is the Newsroom edition of the Sifted podcast and I'm your host Freya Pratt. Late last week, the European tech ecosystem got a glimpse of a future many had worried about. US AI giant Anthropic announced it had been ordered by the US government to suspend access to its most advanced AI models for foreign nationals. The restrictions may prove temporary, but they've caused a big reaction in Europe. It's being described as a wake-up call. Today, we're discussing what happened, why it matters, and what Europe should do next. Joining me to do so is senior reporter Daphne Le Prance-Range.

0:37Hello. Daphne, where are you joining us from today? I'm joining as ever from Paris. Great. And I am in London, where I am based. Let's start then with what actually happened. Talk us through the events of the end of last week. So what actually happened? Anthropic, which is one of the leading US AI labs, the maker of the chatbot Claude, which a lot of us use every day, was ordered by the US government via what's called an export control directive to cut off access to two of its most powerful models to all non-US citizens. So the models this applied to were Mythos and Fable 5, which are some of Anthropik's highest performing models.

1:20They're basically the same model. The only difference is that Fable is more consumer oriented and has more safeguards. And so for now, Anthropic has cut off access to all of its customers out of safety to be sure that it can implement the directive. It's unclear when and if Europeans will ever get access back to these models. What's interesting is that the directive was issued for national security concerns. And so the details of what these concerns are exactly were not disclosed. But Anthropic said the decision was likely prompted by fears of jailbreaking. So jailbreaking happens when users can get around the safeguards of a model, get past those safeguards to carry out malevolent actions, things like accessing private data or prompting the model to give answers to questions that it shouldn't be allowed to give answers to.

2:12So essentially the model is deemed not safe enough by the US government who prompted this export control directive as a result. This is important because it's the first example of the so-called digital kill switch, which has been worrying Europeans for a long time. It's this idea that Europe is very reliant on the US for some critical technologies, AI being one of them, but not only. We've also got cloud services that Europe is very dependent on the US for and other technologies. There's been a lot of worry in recent years that access to these key technologies could be restricted or even turned off by the Trump administration in the context of these escalating tariff wars and so on.

2:56And so Europeans have been preparing for this possibility. They've been aware of their dependencies and their over-reliance on US tech, but it's kind of never happened until now. This is the first instance of this digital kill switch happening in real time and actually tangibly restricting access for Europeans to some critical technologies. I thought it was interesting that the news came a day after a group of quite high profile researchers, including Visionary's VC Judith Dardot, published a paper called Europe 2031, which called on European governments to bolster the continent's AI industry in the face of these geopolitical dynamics.

3:32So it instantly validated their paper. This happened just after they published, which was quite good timing for them, probably. Yes. And also, interestingly, it came just over a week after the EU Commission published its tech sovereignty package, which the very existence of the tech sovereignty package is linked to that possibility of a digital kill switch. And the tech sovereignty package basically attempts to reduce Europe's reliance on the US and, well, on foreign players by that, meaning the US and to an extent China. for some key technologies by, for example, increasing Europe's capabilities in cloud and AI infrastructure and also encouraging European organizations to, first of all, assess their dependencies and, second of all, for critical services actively make the choice to reduce the proeminence of foreign technology providers.

4:26So yeah, again, a very timely event to have come just a few days or over a week after that. Exactly. You've been speaking to both founders and investors about the situation. From talking to them, how existential do you think this is for their companies? So founders and investors obviously are talking about it a lot. I think what I thought was very interesting with this was the political reaction as well. I don't know how it was in the UK, but in France, it came from all sides of the political spectrum. The former prime minister, Gabriel Attel, talked about AI being a technological war that France and Europe must absolutely wage for their sovereignty.

5:03The president of the far right, Rassemblement National, who is not normally super active on technological topics, or at least a bit further removed from those topics than other parties, also spoke about the events. So I thought it was interesting that it kind of across the political spectrum attracted reactions and quite strong worded reactions. Obviously, AI being a technological war is not, you know, it's not a small thing to say. So that was interesting in terms of founders and investors and businesses. I was told it was, you know, the only thing that anyone could talk about all weekend. I was told by one founder that we're entering a new phase.

5:46so it definitely felt like a bit of a turning point. There's obviously a lot of fear around this move but what's the kind of everyday tangible effect on a European startup do you think? Yeah I think the effect can be put into perspective so Mythos and Fable are obviously they're very popular models they're very high performing models but many companies in Europe will be using other versions of Claude which are still available some are using other providers many are even turning to Chinese providers now, which are known to open very high performance open source models. So I guess the impact on businesses is still CBC, but I think it's about the bigger picture, if that makes sense.

6:29So the idea that if this can happen to these models, what else can it happen with? Could it happen with other models, for example, those provided by OpenAI? Could it happen with, as I said at the beginning, cloud providers, which 70 % of Europe's cloud spend goes to three American hyperscalers, AWS, Google Cloud and Microsoft Azure. So that would obviously have a huge impact. So I think there's a realization for businesses, something they've known for a long time, but that is actually becoming a reality that they can't afford anymore to take the risk of relying on these technologies for critical workloads, especially if you're a company in a sector like finance or healthcare or critical infrastructure.

7:10So these businesses need technology that they can rely on that won't be turned off just because of a particular geopolitical situation or dynamic. And so, yeah, this is obviously something that's been known for a while that, as I said, the tech sovereignty package came in over a week ago, as you said, Europe 2031, you know, was this call for action happened last week. So before this happened, before the anthropic restrictions happened, but it's, I think it's made, it's kind of turned that potential hypothetical situation into a real possibility and is prompting a lot of businesses to actually act.

7:45Yeah, you're right. There's a realisation this is way bigger than Anthropic. One argument we've heard is that this could end up being great news for Europe's own frontier AI companies, primarily meaning Mistral, which sits in France. Yeah, as you said, primarily meaning Mistral because there aren't really, apart from Mistral, any other players in Europe that compete or present themselves as competing with frontier AI labs like Anthropic and OpenAI. So Mistral does fill exactly that gap by offering not only a homegrown European model, but also an open source model. And since their launch, Mistral have pitched themselves as that more reliable partner for European companies.

8:29They've been kind of banging the drum for European sovereignty, increasingly so in the geopolitical context ever since 2023, which is when they launched. They've always said dependencies on the US in terms of technology like AI are dangerous. So this event has kind of proven them right. Maybe it's their time to shine. And actually, Ms. Raul are reportedly in talks to be raising a massive round,$3 billion at a$20 billion valuation. So yes, I think we've both been hearing, Freya, that this might help them close that round. But I guess to put things into perspective, that$20 billion valuation compares to a$1 trillion valuation for Anthropics.

9:09So there are questions around, you know, the extent of what Mistral can do and whether it can represent that alternative to American tech. And it's not just Mistral, I think, just in general in Europe, there is that question around whether the alternatives are there. Yeah, I think it could give a boost to a number of companies. We've seen, for example, Lovable, which is a vibe coding company from Sweden, talk increasingly about their relationship to sovereign Europe. So I think even the ones that aren't providing the frontier models themselves might see this as a moment to push themselves forward.

9:40I spoke to this week a researcher at Hugging Face, which is another French startup, a guy called Yassine Genite, who said that although he recognized the seriousness of the situation, we need to remember that for many AI tasks, a smaller model is as effective. So perhaps this thinking that we rely so heavily on anthropic might need some nuance to it that actually we could, for everyday tasks, in many cases turn to smaller models, which Europe has more of, which I thought was an interesting angle to the debate that I hadn't seen as much in the kind of initial fearful response to anthropics moves.

10:16Definitely. And I think there's also a cost concern. So these big generalist models like Anthropics are very, as I said, high performance, but they're also costly. And it just kind of backs your point, Freya. I think there is a move towards using smaller, more efficient, less compute intensive, less costly models that aren't necessarily the huge generalist models that are pushed by the big AI labs that we know from the US. yeah potentially there is a gap for that in Europe at the moment. Exactly. What have founders and investors you've spoken to called for from governments in their response? I know there's talk from some founders of Pause the AI Act for example.

10:57Yes we've heard calls to pause the AI Act and I think more generally calls to move away from a perceived over-regulation from the EU. I think generally what everyone can agree on is the right responses to create European champions, not only in AI, in other technologies as well, in cloud, in semiconductor. For some, there is a lot of grief with Europe's tendency to overregulate, which they see as a barrier to that. So even the European tech sovereignty package, which was intended to enable Europe to create its own champions, even that was perceived by some as just an extra bit of regulation. And I think there was a bit of skepticism with the idea idea that publishing yet another set of proposals for new laws and new rules would be the way to create these champions.

11:47So yes, there's been calls to pause the AI Act. That's another text that's a bit older that was passed in 2024, which regulates players that are making AI models and also players that are implementing and deploying AI models in their businesses or for use cases and that has caused a lot of controversy in the past already. But yes, some founders and observers are jumping off this new development to call for less regulation towards the AI Act so that Europe can race forward and let the potential tech champions become exactly that. Another thing I think that a lot of people will be keeping an eye on is the G7 summit that's happening in France this week which obviously is not specifically tech focused but certainly from the founders and observers I've spoken to expectations are high that the anthropic development will be a topic of conversation and that there will be a discussion around that and presentation of how how we could go forward as the EU as a bloc.

12:49I spoke to Judith Dada who I said talked about earlier as the co-author of that paper on the AI future for the continent. She sent over some thoughts after the anthropic ban and her idea is that Europe should form a kind of alliance of middle powers, she called it. Middle powers being those that exert significant diplomatic and economic influence but slightly lower than something like a superpower I've seen in America and China. So her idea is that countries like those in Europe, but also Japan, South Korea, Canada should come together as a sort of aligned coalition to work on their AI strategy.

13:26South Korea and Japan being really useful because they're integrated within the chip supply chain, for example. So I thought that was interesting, people starting to think outside of the continent, who do we need as allies at this moment? Yeah, definitely. And I think that's been also in the making for the past few years. So, for example, the AI Action Summit that happened in Paris in 2025, it was very much about trying to find that support within other regions that might be strong partners to develop, to kind of break away from that US and China dominance and see if there was like a third way.

14:01And again, these are things that have been in the making for many months, for years even, and as preparation for what was always a hypothetical situation. And I think now it's happening, it might accelerate things quite significantly. Yeah, it's an ongoing debate and we'll keep tracking it as there'll be twists and turns. I'm sure that the ban will move in some way and we will be reporting on what that means. That's all we've got time for today on the topic. But Daphne, before we go, tell us what you're working on, what listeners can look out for from you. Well, this week is Viva Tech in Paris, which is potentially the biggest tech event of the year.

14:36There's always a lot of announcements. I think the president is expected to come along. My understanding is there's going to be a quite strong focus on the Franco-German relationship, which kind of goes to what we were saying and to the need to bring forces together in order to ever have a chance to create a credible force that can rival the US and China. China. So I'll be keeping track of those announcements. If you fancy grabbing a coffee with myself or actually with my colleague Maya, who will be there as well, feel free to send us an email. My email is daphne at sifted.eu and Maya's is maya at sifted.eu.

15:17I'm also speaking to Domin, which is an Italian model developer because I'm interested in their thoughts on the anthropic ban and what they think the way forward should be. So yes, stay tuned for that. Great. And you can catch Daphne's reporting along with Sifted's wider coverage of the ongoing sovereignty debate in our newsletters, which we'll put some links to in the show notes for this episode. As always, please rate, review and share the podcast. And this episode was expertly produced by Maya Durampo-Hornby.