Anthropic vs European AI

0:02Hello and welcome to the Sifted podcast. I'm Freya Pratty, Associate Editor here at Sifted. At Sifted, we spend 99 % of our time writing about European tech companies. That's our focus. But across the last few weeks, we found ourselves spending a decent chunk of time writing about an American one. Anthropic is the US model maker behind AI tool Claude, which lots of people now use. And a number of its recent moves have made big waves in the US and also in Europe. It's, for example, planning a big expansion in the UK. It's got this office space lined up for 800 people in London. It's also, and this is the most talked about part of its recent moves, it's been releasing challenger products to compete with some of Europe's busiest startups.

0:44So joining me to talk us through Anthropix impacts on European tech is my colleague Martin Coulter, SIFTA's News Editor. Martin, thanks for joining us on the pod. Thank you very much for having me. First off, give us a bit more background on Anthropix. So Anthropic was basically born out of open AI drama before it became mainstream entertainment. The founders, particularly Dario and Daniela Amodi, this brother and sister team, were senior people at open AI who left partly over disagreements around safety and commercialization. They kind of position themselves as this grown-up AI lab, very focused on safety, very research-heavy.

1:29A lot of talk about guardrails and constitutional AI. Claw, their kind of chat GPT equivalent, became the safer chatbot. And now, you know, as they've grown rapidly over the last year or so, they've moved into coding, finance, legal work, agents, and have become much more of a challenger to open AI than I think anyone ever really expected. I think the moment when we as a Sifted Newsroom first started writing more about them was Feb this year. They put the legal AI industry on a bit of a defensive. Can you explain more about the release and what that triggered? Yeah, so this is part of the wider SaaS-pocalypse where there's this idea that AI companies, through intuitive coding access to simpler no-code solutions, are going to be able to create challenges to these big incumbents very quickly.

2:29The legal thing in particular feels like one of the first genuinely scary AI moments for big European and American companies. With this, Anthropica has basically gone after the really lucrative grunt work that lawyers bill for contract reviews, NDAs, compliance checks, all that kind of stuff. And the reaction on the public markets was pretty brutal. You saw this sell-off of Relics and Pearson, various companies working in these areas because investors suddenly realized this isn't just AI assistant territory anymore. It was a big shift. AI isn't just helping software. It's actually starting to replace some of this software.

3:08What I thought was interesting was the markets reacted really strongly. But when you looked at the actual update, it was some prompts specific for the legal profession, right? So maybe I was wrong. I didn't think the changes in themselves were like huge, but I guess it was the potential that it showed a very rich model maker can just slam into a young industry. And the potential of that move to go further into them was my read on it. Because when I looked at the prompts themselves, they didn't seem to be going as deep as something like Swedish AI firm Lugora, which is working on legal tools. And we've covered a lot.

3:44Is that kind of what you'll read on the situation? I think so. I think there's an extent to which everyone is wondering, and this is like the public markets are driven primarily quite often by certainty versus uncertainty. And when a company like Anthropic releases these tools, which on the surface look like they could, or at least they claim will fundamentally upend a certain industry, you know, the price is going to go down and people are going to worry that maybe this is going to be something that's going to take a hit. so i think you know there has been a recovery of sorts since then but yeah is it that big a deal it's very hard to know like well you know a company will lose millions of dollars one day and then recover that over the course of a couple of weeks so tbc right and it did represent that their ability to move with load of money really quickly right yeah the interesting thing i did a bit of chatting to the european legal ai firms in the aftermath of that release and they They said what I think we thought they would say.

4:45Lagora said that they are actually built with the reality of legal work more in Trent. So it can integrate with Word and Outlook. And they've got teams that you can go to for support. So I think those firms are trying to really sell that they are deeper embedded in the industries they're working on. The other thing that they really tried to sell to me was that some of them are verticalized. So they're focused not on law as an entire industry, but on real estate law. So they've got that deep expertise, which I thought was interesting. and I think that for the publicly traded older companies they're going to bill themselves as owners of data and therefore better trained models is the impression I got.

5:20Yeah they're all looking for a way to sort of defend themselves. The legal sector seems to be the one definitely on the public markets has kind of had this big challenge up front and they're still working it out. The question now I think is whether this is the first domino and whether or not various Are there industries that are going to be upended by this? Or whether or not it's just another part of the famed AI bubble. And actually, there's nothing to worry about. But we'll wait and see. Yeah, from talking to investors at the moment, it feels like everyone is in the let's wait and see mode, which is perhaps not helpful, but maybe the best they can do.

5:58The other thing I thought was interesting was they're not – Anthropik isn't the first one to launch a kind of verticalized offering, right? I think OpenAI have also released ChatGPT for health, but it didn't seem to have as kind of bigger hits on public markets or investors, which I thought was kind of interesting. But the other thing that Anthropic have done that has changed the sifted newsroom agenda recently was the leak about them working on a vibe coding product similar to Lovable. How do we know about this one and what do we know? So, yeah, there was a leak online which was doing the rounds of a vibe coding platform that Anthropic was apparently working on, which very much seemed to be being positioned as this kind of challenger to Lovable.

6:40Obviously, Lovable, as we all know, is one of the most hyped startups in Europe. Their annual recurring revenue has been a big talking point over the past year. They've drawn a lot of investment. I think they fundraised twice last year, huge amounts of money. And Anthropic is apparently testing this in-chat app builder where you basically just describe an app and Claude will build the whole thing, which is exactly what Lovable does. Lovable's huge now. It's got a$6.6 billion valuation. So the stakes are massive. And I think this is becoming the kind of defining anxiety for European AI startups.

7:18Are you building a real company or are you just building like this temporary layer on top of someone else's model? that they can then come out and copy. And to Lovable's credit, I mean, they've moved quite quickly. As yourself and our colleague Maya reported on, they released this desktop app version of their product very quickly. And I think they're feeling the pressure to kind of respond and work out what their next move is. Yeah. So I am in contact with an employee there who talked about, as you said, they released this desktop app. And this employee said to me that they'd worked through the night in response.

7:55I think they must have been working on it for a while, but to accelerate the release. So they were clearly aware they needed to respond in some way, which is interesting. I thought your take on the sovereignty angle that founder Anton Ossica then came out with was really interesting. Yeah, this is a new trend, which we're seeing a bit of. Basically, whenever one of our most promising AI startups comes under a degree of pressure from any angle. I think we've seen it with Arte Mench at Mistral AI in France. And we saw a similar thing with Anton Azica very shortly after this story broke, posting about European sovereignty all of a sudden.

8:33And I think that's becoming more and more of kind of a key part of the pitch for European startups as they become more and more aware of the challenge posed by the far bigger American companies. Yeah. And I think investors react to that language, right? I wonder if consumers do. I feel like you just look for the best tool, right? You don't think, are they Swedish or are they in California? I think you just look for the best tools. So I wonder how much cut through that will have on user uptake. But you're right, we're seeing it way more than we were. The other anthropic move that we've been thinking about, the company themselves flagged the danger of their new model called Mythos, which they said is so powerful it can exploit unknown issues in big systems.

9:16and that's left banks and finance ministers asking for access to see if they can stand a chance of securing their systems. How are Europe's fintechs, because we do now have some big names, how are they thinking about that threat, do you think? Yeah, I'm always aware. I never actually really know what the word powerful means in the context of AI. I don't know if that just means they can do impressive things very quickly or if they're like, I feel like we don't talk enough about existential risk and the AI apocalypse much now, but that was like the concern at one point. But with Mythos, this model developed by Anthropic, it feels like the conversation has suddenly got quite dystopian again.

9:59This isn't just a chatbot that can code, as we understand it. It can apparently autonomously find and exploit kind of cybersecurity vulnerabilities at a massive scale. and regulators are freaking out because they're treating it less like a product launch and more like a potential financial stability issue. The fear is basically AI-powered cyber attacks could happen faster than any institution could hope to defend themselves. And the kind of controversial bit is that governments and, you know, chief execs and CFOs and cybersecurity leads at some of the biggest institutions have all been getting together.

10:42They've all been demanding kind of early access to mythos, engagement with Anthropic to try and understand how it works. But also, you know, in the event that this was used by a bad actor or, you know, somehow ran amok, you know, they want this kind of priority access so that they can best understand how they can respond to it. But it looks like at this stage there may be a kind of two-tier system emerging early on where you've got JP Morgan or one of the other big investment banks, et cetera, kind of getting into the room because they have this access and they have this huge company already embroiled and with the right contacts, getting to patch itself early, getting to be in those conversations.

11:27whereas smaller fintechs which you know are still growing but don't necessarily have those kind of connections may risk being exposed in the long term it's especially awkward in europe because there's so much fintech innovation happening here and smaller companies just don't have those relationships don't have that access so the people moving fastest on innovation may actually end up being the most vulnerable. Do we know of any European fintechs with access or communication with Anthropik? We have asked. We understand that they are asking for it. Whether or not they're getting it is a different question.

12:06I also think, and I think Anthropik are good at this, this is amazing PR, right, to have headlines everywhere saying this model. So I'm sure it is powerful, but it's great PR for their company. Yeah, yeah. No, it's a cyclical thing and something that I think we try and always be aware of with these massive claims. And I think this kind of talk peaked in, was it 2023, when we had the AI Safety Summit with Rishi Sunak at Pletchley Park? I think it was 2024. But that kind of, it's great PR, this idea that our AI models are so powerful and terrifying that they require a kind of global summit. I mean, the inevitable consequence of that is, you know, you should deregulate us and you should help us secure as much money as possible because we're the only ones who can help defend you from the things we're building.

13:00I've never quite got my head around it, but yeah. Maybe European tech needs to go more down that route. One European company we think has a relationship with Anthropic is Fractile. Talk us through what we know about that potential deal. Yeah, so this is a deal that's been reported that Anthropic is apparently looking to buy AI chips from Fractile. This is a company that I believe originally spun out from the University of Oxford. We've done a fair bit of reporting on them in the past. I mean, the basic underlying story here is that Anthropic wants alternatives to NVIDIA, one of the biggest companies in the world that makes chips that almost the entire AI LLM ecosystem runs on because everyone's completely dependent on NVIDIA right now.

13:54And I think that is seen as a kind of long term risk. Fractal's whole pitch is basically their inference chips are dramatically faster and cheaper. And if that tech works, it could matter a lot because inference, the process of actually running these models, is an incredibly expensive part of the business. So it's a huge kind of validation for UK deep tech. But there's a kind of strange underlying story here, you know, which comes up again and again in our reporting about Europe providing, you know, some important components, part of the infrastructure. But U.S. companies still owning the platforms, owning the models and, you know, just being so much bigger than us in so many ways.

14:42I think it's also good validation because Fractal is one that's name checked by people like the AI minister and the government. it's good validation that they are kind of hyping up ones that are actually sought after by the market was my read on it but fractal is super interesting i hear constantly how enthused investors are about them so i am working on a profile at the moment of their founder walter goodwin who seems to have a great reputation in the market so if people have thoughts on the company or him please get in touch well i'm thoroughly looking forward to reading that thank you and if the topics in this pod have piqued your interest you can sign up to our ai and deep tech newsletter which covers this topic as well as our daily newsletter which goes out daily and often finds itself covering anthropic at present you can find a link to both in the episode description as usual please rate and review the pod it really helps us reach new ears and work out what people want to hear about and this pod was produced by tim smith