The Connected AI podcast: Anthropic's head of Europe speaks on trust, safety and the AI race

The Connected AI podcast with Charitator and Elaine Burke in association with PartsOn. Hello and welcome to the Connected AI podcast. This week we have a special one for you. I am going to be talking to Guillem Prinson, head of Emea for Anthropic. Now as you know Anthropic is one of the biggest stars in the AI world right now. A close rival to OpenAI. The company was founded by the Amodes who left OpenAI to form their own company because they were worried about safety and wanting to bake safety in ahead of everything else. Now we have a great conversation with Guillem covering everything from potential IPO, problems with the Pentagon, how Ireland is doing in terms of AI and much, much more. Stay tuned and I think you're going to like this one. When the stakes are high, clarity beats guesswork. PartsOn helps finance, law, security and regulation assess risk fast, using intelligence they can trust. By bringing complex data into one key review, PartsOn covers hidden ownership, conceit links and risks others miss. For clarity you can trust when it matters most, visit PartsOn.com for more information. Well come here, welcome to the podcast. Great to have you with us. It's hard to know even where to begin with you because I mean Anthropic is so busy at the moment. There's so much going on with the company. It's overwhelming I imagine. But let's start with Ireland because you just recently announced a load of new jobs for Ireland. I suppose the first question that pops in my head is why Ireland and why now? Yeah, thanks for having me, Charlie. Listen, about Ireland and our position here, Ireland offers a very deep AI talent pool on the one side and also a pro innovation government to cure our expansion. Maybe just to say a few words about the government side of things, Anthropic. Absolutely. We're very proud to be working with the government and the IDA to support Ireland's growing status as a global hub for AI. I think the Irish government has delivered its own digital AI 2030 strategy, which sets out Ireland's ambition to become a global hub for AI and build on the country's great tech talent base. The other thing I'm seeing in Ireland is, I mean, we are seeing is there are amazing companies. I mean Stripe is one coming out of Ireland. But also some leading Irish startups like Tines and Wayflier and Manna, they're all called users by the way the ones I've just been to. Coincidentally. I was pretty coincidentally. Why not? And listen, these Irish startups are using cloud to solve real world problems. Another example is Zero Stripe. I started founded by Irish university students that recently won a first place at the major European coding competition with an idea to prevent wildfires before they start. So we're seeing lots of startup activity in Dublin and in Ireland generally. And then the more general picture also about Ireland and how it's picking up AI beyond startups that are doing so obviously natively is that we regularly publish our anthropic economic index, which is a publicly available on our website, economic index, that kind of reflects the usage of AI across companies and across countries in the world. Because we believe we have this responsibility to study and report out on the real world impacts of technology that we're developing. And this is a way to give early science of what's coming and helps people say how they want the world to look like. And when we look at the country analysis, what we're seeing is that Ireland actually has an overall very high usage of AI and anthropic in particular that ranks Ireland as the 16th, 16th country out of the 116 countries we do this kind of analysis about. This actually puts Ireland ahead of other large European economies like Germany that ranks only 28th. And so I think there's something beyond the government backing, beyond the extremely good talent pool, beyond the vibrant startup ecosystem that's building AI native companies. I think there's also a general acceleration in AI adoption and usage across the Irish population. And of course, you're looking after Emea as a whole. And you had a similar job when you were with Stripe, which we always love to plug on this. So what is it like because anthropic is growing so fast? What is it like in comparison to when you were in child of scale in Stripe? I mean, is it a similar story for you where you kind of go, "Ah, yeah, I've been here to do this before, or is it, of a magnitude that's kind of completely different?" Yeah, I mean, listen, yeah, thanks for reminding. So I spent over seven years building and scaling Stripe across Emea prior to this and anthropic. And yes, there are similarities. I mean, one is, Stripe is also in the businesses of trust andthropic is in the business of trust. AI usage is a business of trust and we anthropic make it the priority in the way we develop the company and in the way the company was, by the way, founded a few years ago. Another aspect that's very similar is anthropic is a platform company. We build LLMs, the most powerful LLMs of the world, as an infrastructure for other companies to use or to build on. Kind of in the same way, Stripe is a payments infrastructure that other companies use to get paid and to do money transfers. And so that's another similarity. And I think thinking about how different it is, one of the things that really comes to mind is the speed of adoption, frankly, of this AI technology. And I'm happy to say, much more about this across Europe in particular. But yes, we're seeing an anthropic, we're seeing AI being adopted in a way faster way that I think any technology beforehand. And when you look at our, for example, the anthropic revenue has more than 11x in the past year. That's incredible. And the trajectory we're on. And we started as Dublin being our first office in the EU in 2024. And we're now a five office triple digit headcount European operation. With again, as I mentioned, 11x here on your revenue growth in Amia and customers across the continent. And that trajectory tells you everything about the European demand for AI they can trust. And obviously our Dublin presence is a key element to that. And just maybe to hold on that for a second, because you mentioned this a little earlier is, yes, we're super proud to have recently announced an increase in our office space to 21,000 square meters here in Dublin. And we're hiring over 200 people across sales engineering and finance by the end of the year. And we're a Dublin office, close to 300 people by 2027. And that's something that's very exciting to us. Yeah, that's fantastic. I mean, there was a lot of excitement when the news was announced as well. And tell me, I wanted to go back to the sort of start of anthropic because I mean, you touched on it there. Or for you to trust is, is critical both to strive when it was starting our own on its journey and still is and is obviously very much a case for anthropic as well. Now, anthropic started with, you know, the Omodis leaving open AI and forming the company because it was a lot around safety issues and then wanted to kind of be a safety first company. Could you tell us a little bit more about that in terms of, you know, I've interviewed people in the company before, like Mike Krieger, who was the chief product officer, who was saying, you know, it's not about speed. Obviously, everyone's in a rush to kind of, you know, dig in on this, but it's not about speed at the cost of safety. I suppose if you could touch on that of like, is safety still paramount to the company as it once was? I mean, you laid it out very well, Charlie. Safety is that the DNA of anthropic. That is, as you mentioned, that is, that is the reason why the company was created for five years ago. Anthropic had a safety team even before having a product. Anthropic is baking safety at the core of how we are building product. For anthropic, safety is not an after fact of having built a product. But you know, two examples of how that translates in real life. One is we have what we call an RSP, which is a responsible scaling policy, which essentially is a policy that dictates all the things we want to have right before releasing new models on the market. And yes, if it takes a bit more time to get those things right before releasing models on the market, then that's a chance we're going to take. Because we believe it's paramount to get safe models out on the market, especially as these models get increasingly powerful. A second example is the concept of constitutional and I that we created, which is to say that we've defined a constitution for our AI models. This constitution, by the way, has evolved over the years and we released our latest one a couple of months ago. And this constitution is a way to put some guardrails around the way models think and respond to questions. And so that's another way of kind of putting guardrails around how the models behave. And so with all that, what we're trying to do is we're trying to pioneer what we call a race to the top on responsible AI, prioritizing compliance and upholding privacy and security standards. That by the way, European companies actually are very much in demand of even more so than elsewhere. And so if I just ask, is that partly because safety is baked in perhaps with EU regulations? Because obviously that's a question I want to talk to you at some point. as well, is that sort of part of that or is it just other reasons? I mean, yes, Europe sets a very high bar for AI safety and that's exactly the one we're trying to meet and we want to meet and we actually want to push the industry up to. That's what we call the race to the top. And that is also why we are the prime partner for highly regulated industries like healthcare and financial services and legal and government. That's why they come to us. They see us as that, you know, trustful player and they also want to, frankly, in that sense associate their brand to that level of trust that we showed. Now, I think the important thing here is, Stake does not mean slow. And I think what my point of view, what I'm going to tell you is our goal in the end is to prove that Prontira AI can be developed safely in responsibility and responsibly, sorry. So Prontira AI can be developed safely and responsibly. We believe that safety can be a competitive advantage and not a burden. And that's the whole idea of it. Like we cannot be safe and responsible if we're not at the frontier. And guess what, we are at the frontier and we are the safe and responsible player. And that's the, that is above all priorities, the priorities of anthropic as a company. That's a really interesting point. You say about it giving you a competitive advantage because I mean, that's one thing we're seeing. And you know, there's plenty of reports out about how well anthropic is doing in terms of bringing revenues in how close it is to profitability versus other AI companies. And you know, I mean, in that, I mean, you seem to really be scoring, particularly with businesses. Could you talk to a bit on that? I mean, is it that competitive advantage that's bringing you this added income coming in? Or is it also obviously that how good the quality of the product are? I mean, listen, we've never been shy to say that our focus is on the enterprise. Anthropics focus on is on the enterprise. Today, 80% of our revenue is enterprises and not the other way around. And so we are building our products for the enterprise. And so if you kind of take a step back and look at, you know, take the past 12 months, right? And look at the evolution of the products we've been shipping and our focus, you know, especially like, including if you look at Europe, Claude is basically a Claude on anthropic or changing how Europe is working. They started with developers, but then now it's getting to everybody. And I'll say it worked here. It started with developers and Claude Cod went from being an internal tool at an anthropic a year ago that we launched to an over 2.5 billion business and revenue run rate, you know, a year after launch. Six months after launch, it was already a billion dollar revenue run rate. And this is about, you know, redesigning how engineering teams are working. And European engineering teams have been a big part of that, right? We've been on the ground with them, Spotify, Insta-com, Lovable as well as Stockholm is building entirely on anthropic, station F in Paris, and bigger companies like Allianz, right, in Germany. They're rebuilding, they're redesigning how they're engineering teams are working, thanks to Claude code. Then, after Claude code, a few months ago, we launched Cod. And Cod is basically bringing to any knowledge workers what Claude code was bringing to developers. And so you're basically expanding this transformation to all knowledge workers across the world and across Europe. So reporters, analysts, accountants, consultants, without the command line. So, image capabilities, but no technical barriers. And that's the transformation we're seeing. Now and now, and even further, now in people's pockets. So now, Claude, Claude has hit the top, you know, app store rankings across Europe this year. Yeah, it's become increasingly popular with consumers, partly obviously, as a response to the stuff going on with the pentagon. But suddenly, like, consumers are going, I want this as well. So you're winning, I know, a no segments. Because they're also, you know, they're discovering Claude at work. And then you want to use it for personal. How are the other way around, by the way, they want to use it personally, then they bring it to work. And by the way, including in Ireland, that kind of ranking in the app store. And so they're choosing an AI, you know, they work with for usefulness rather than stickiness, right? People don't want to be sticky. They want the AI to be useful. And so we're starting to break through into this site guys and give people a healthy choice in AI. Claude, as I mentioned at the beginning, is for people who want AI, they can trust AI that expands their thinking rather than replacing it. And that's built to benefit them not to capture their attention, right? That's really how we think about what we're building for the enterprise. That makes sense. I've seen huge reaction to co-work in particular and obviously Claude, you know, the code in is just unreal, you know, what it's capable of doing. But there have also been concerns raised, I suppose, you know, we've had the Sasspocalypse recently where which was largely on the back or partly on the back of Amphropic bringing out new tools that made the market kind of go, oh my god, Amphropic is kind of going to come and eat all our dinner. Could you touch on that? I mean, there are fears, you know, of Amphropic kind of, I suppose going beyond what we've come to expect from NLMs and then kind of going after which would have been secure tech businesses. Yeah, listen, I mean, listen, I'm not here to comment about, you know, market, social kind of stock prices, but what I will say is Amphropic builds foundational AI models. That's what we're building. We're not building vertical applications or law firms, for publishers, for educational institutions or even like type Sassp businesses, like that's not what we're building because the real value in these industries lives, does live at the application later. You know, an application built for, I don't know, accountancy, for example, has decades of domain expertise, proprietary data, you know, deeply embedded customer relationships, compliance, we don't replicate that. Like others do that much better than we do. Application layers do that much better than we do. So by building with Claw, those companies, those application companies are combining the best industry knowledge and expertise. There's with our safe and powerful models to create the best outcomes for end customers that, you know, only they could do, right? That is really our positioning. And so, yeah, back to that infrastructure and platform layer that we're building to enable others to build applications or transform their businesses based on our platform. Does that mean people shouldn't be afraid that Amphropic is going to come and eat it? I mean, I think people should make the most opportunity of having access of these, you know, powerful tools that, you know, didn't exist in humanity even, you know, six months or a year ago, right? And I think the world has never been the world of opportunity for anybody wanting to start a company has never been wider, right? Has never been greater. And I think that's more the message to kind of walk out with, which is now's the moment to start those companies and to potentially, you know, challenge some incumbents, right, on the application layer because you have such powerful tools available to do so. I mean, I wanted to just start back again on the, I suppose the relationship with OpenAI, I mean, you know, Amphropic was born from OpenAI. It went out and it's done its own thing. I suppose when it started, you know, OpenAI was kind of seen as the leader in the space because it was out very early with chat TPP. But now, you know, yourselves Google with Gemini are out there and, you know, some people would say you're doing a better job than OpenAI was, you know, outside of that though, I just wanted to touch on the rivalry there with OpenAI and with other competitors in the field. Do you pay much attention to what's going on to them? Is it very much a case of like, we need to be better than them. We learn from them or what is your approach to rivals? I mean, we're very focused. We're focused on delivering on our priorities first and that is twofold. One, we're focused on building the most capable, reliable and safe AI systems. Our benchmarks show cloud continues to lead encoding and and agenda capabilities. The areas are enterprise customers really care about. The last point, one point, two is we're an enterprise AI company, as I mentioned before, not a consumer tech company. 80% of our revenue is enterprise. 300,000 businesses across the world. And I think you've seen some symptoms, more than symptoms, some illustrations of that in the past weeks, really differentiating our positioning on that. Like, we don't need an ad model, for example. We're not pushing ads on cloud for many reasons I'm happy to talk about. Like, this is something we are able and wanting to hold hard on because again, we have this enterprise focus and we want to be focusing on building for the enterprise. And I guess it's a very tight focus you have, even though you're expanding by the day. I suppose compared to say, open AI, they've now ditched Sora. They've got the relationship with Johnny Ive and who knows what happens with that, whether they're bringing out a product or any of these things. You've been sort of, I suppose, going after everything, whereas you've been very focused. Is that being of huge importance to just kind of go, look, let's win in the enterprise space and let's not get sidetracked anywhere along? We've been very deliberate about focusing and building for our enterprise customers. Yes. That's a good way of responding. And tell me, because I just want to go back as well, because you mentioned about Europe and you're talking about safety. Obviously, there's been plenty of people, tech bros around the world, kind of going, "Boo!" to a regulation that's slowing us down. But, you know what I mean? It's got, I suppose, refreshing to. you say that you're looking up to the regulation here, you're paying attention to what's going on in Europe in terms of regulation. The EU has stepped back a little bit in recent weeks and months in terms of bringing in regulations on AI. Do you welcome that? Is it a nice balance that can help to work out that line between innovation and regulation? Listen, our review on regulation starts from the position that Dario laid out in his latest essay, which is called the adolescence of technology. That position is the following. Some government action is necessary, but it should be as simple as targeted as possible, starting with transparency requirements and evolving based on evidence rather than ideology. That's really where we're starting from. So we believe in regulation. I think transparency requirements is really the thing we believe is the most important. We're working with regulators across the world, including in Europe and in the EU to push those messages. As we've signed, we were the early signers of the EU's color practice and we engage with those regulators very regularly. So we believe in the value of regulation and we're in discussions with regulators around that to make sure we're doing the right thing. I think what's important with our approach is what we're trying to avoid is the move fast and break things approach. We've seen that happen in the past in various other areas of the technology world. That's something we don't believe is right in this era of AI given the power of this technology. So it's all about making those calculated moves that benefit our customers and society. We prioritize protection over engaging metrics is how I would say it. And guess what? Customers value our mission and these ethics. They trust that they can align their brand to ours in that way and that's important to us as well. Yeah, that makes sense. And talking of regulations or the impact of the political space, let's talk about the Pentagon and their moves against the Amphropic recently. I know you may be limited in what you're saying because there's stuff going going with that. But I mean, has it been damaged in for business? Are you able to work around whatever you get for owner in terms of from the Pentagon, whether you're a threat to the system or not? So a couple of things. One is just to be very clear. There's a CA category that the Pentagon has put us in only limits a very slim amount of business that's been, you know, confirmed by our cloud partners, Google, Amazon and Azure, which only limits to direct applications for the department or which is a, you know, a slim, a slim, a slim piece. We're what I can say is we're an active litigation. So there's not much we can say here. However, those recent events have catalyzed more serious conversations about AI safety. And that's a good thing. Our business is coming out of it stronger than ever with continued enterprise momentum and people voting with their feet downloading cloud around the world. Yes, a strong strong message being sent there by consumers at least, you know, as well as businesses. Can we talk about IPOs? You know, we've had SpaceX just announced their their filings for IPOs. IPOs seem to be in favor again. We'll also see an open AI and others being talked about we've go in public. What's the story with you guys? I mean, listen, it's fairly standard practice for companies operating in our scale and revenue level to effectively operate as if they were publicly trained companies. We haven't made any decisions about whether even whether to go public and don't have any news to share at this time on that topic. I guess just on that, you know, do you think there would be pause for four in terms of going public just for the fact that as you said, the safety is so paramount when you get into the cycle of your the public market where you've, you know, reporting on a quarterly basis. I know say the Collison brothers have talked about staying private has allowed them to not get caught in that will. You know, is that something that you would be concerned about possibly in terms of going public? When I'm communicating on these kind of things right now, I can't really stay more about these kind of sensitive topics. That's very. So I put you on the spot there. Tell me about, you know, what's, you know, I suppose you've already talked about islands in terms of AI. What's happening I suppose, you know, on a bigger level, you know, when AI first appeared, you know, we're obviously AI has been around for many, many years, but you know, obviously in more recent years where AI is kind of really come to into focus, there's been lots of concerns of AI, is it going to bring us to this new utopia or is it going to destroy humanity? Now, obviously, somewhere in between is probably where society is going to end up, but I just wanted to just talk to you about that and get your sort of sense of what will AI bring to society both in terms of the good and the bad. You know, I mean, we all really witness in huge sea change in things from AI. I mean, let's say I don't think, I don't think anybody has the answer to all that yet. That being said, we, and we certainly don't have it as a unique entity. However, we believe we have a strong role to play to feed, you know, various constituents of society, whether it's academics, economics, politicians, policymakers, corporates, obviously, CEOs, et cetera, with what we're seeing from our angle here. And so our angle is essentially what is the usage of AI? What is it used for? Where is it used? How intensely, et cetera? And again, back to what I mentioned, I think at the beginning of our conversation, we're to do so, we're publishing this anthropic economic index every few months. The last version came out a few weeks ago, and it's publicly available on our website. And essentially shows the usage, like various, you know, various angles around the adoption of AI across the world, across jobs, different types of jobs, different types of tasks, et cetera. I think one of the interesting observations from the last version we released is when looking at, we compared what the theoretical adoption of AI could be among certain tasks and jobs, and what the actually observed adoption of AI, how that compares to the actually observed adoption AI is across those same jobs. And the overall conclusion is that we are still right now at the very start of AI adoption across tasks or jobs that have been identified as potentially using AI a lot. So this is still the beginning of the journey. We're getting early signals about all that, but it's still early signals. Yeah. And as the models get stronger, though, and we're seeing the move in terms of things getting better and better at a phenomenal rate, it's kind of hard to keep up with at all. I mean, are there sort of strategies, I suppose, that Europe, that islands could be taken that will help shore up things for the future? I mean, I think the way we look at this is, yes, AI is going to be transformative for our economies. And in particular, if you're a country, if you're an economy in Europe, I think the one of the things that should be top of mind is how do we capitalize on this new technology and how do we make sure our companies and economy are ready to adopt this technology fast in order to stay to stay in the global race, right? And that's what we're focused on when looking at Europe, right? We're committed to an open and economically competitive Europe with assured access to Frontier AI. And that's the commitment we're making to the countries we operate in and are pushing are helping to support. And especially around the angle of in order to remain competitive, companies need to have access and need to adopt this Frontier AI for scientific and economic progress. So, yes, that's I think a very important angle that various economies are looking into. Last question then, you know, just around things off is like, what's the future look like for anthropic? Obviously, we're not going on, we're not touching on IPOs, but more broadly, you know, in terms of what's coming down the line in terms of products, solutions, and how you see people adopting it and using it in the future. I mean, you know, I'm going to be I'm going to repeat my message around focus, right? We are we're very focused, we're focused on building for enterprise businesses. And so we're going to remain focused on building building the products and the real life kind of agents, the agetic products that enterprise companies are asking for. You know, Cloud Code has been the fastest growing software product in history for the past year. Co-work is an extreme demand right now across all kind of knowledge worker industries and you can expect us to continue to invest along those lines and along kind of real life agents for companies to help them transform all that in, you know, putting AI safety and trust at the center of what we do because that is well is going to remain our long-term focus. And with that focus, does that mean, you know, are you winning the AI race or is it not a race? That's my final final question. I mentioned before I think we're still at the beginning of it. Right? There's still so much to build and yeah, we want to be, we want to be powering the enterprise companies in that way to help them, you know, adapt this technology and make the most of it. Great stuff. Come here. Thanks every so much for taking the time out to chat to us. Brilliant conversation. I look forward to cutting it up again soon. Thank you Charlie. Thanks for having me. Thanks. Bye. Bye.