[0:00:04] So this is a big week for Anthropic. Yes. Your first Super Bowl ad. How do I communicate better with my mom? In its first ever Super Bowl spot, artificial intelligence leader Anthropic is vowing to keep ads out of conversations with its AI assistant, Claude. Great question. Improved communication with your mom can bring you closer. Or if the relationship can't be fixed, find emotional connection with other older women on Golden Encounters.
[0:00:33] The mature dating site that connects sensitive cubs with roaring cougars. What? A tongue-in-cheek approach, making a serious point as competitors like OpenAI begin testing ad placement in AI conversations. OpenAI CEO Sam Altman responded to the ad on X, saying, We would obviously never run ads in the way Anthropic depicts them. We are not stupid, and we know our users would reject that.
[0:01:02] You know, this really isn't intended to be about any other company other than us. And to be clear, you know, our view is not, you know, all ads are bad or there's never, you know, the right place or time for advertising. It felt to us like AI conversations are different, right? People are sometimes uploading private or confidential information to their AI tool. And to us, it just didn't feel like the...
[0:01:31] A respectful way to treat our users' data. Anthropic president Daniela Amodi co-founded the company with her brother, Dario Amodi, in 2021 after leaving OpenAI. What made you leave and when you left, did you have the intention of starting something in the same industry but different? I really felt at the time like we were running towards something versus running away from something. We had this vision in our head of...
[0:02:00] Building an AI company that was transparent, that was really focused on safety and the public benefit, that's why we chose to incorporate as a public benefit corporation to sort of have that really baked into our DNA. And it's funny, it's like when we were there, there were a lot of activities that we were doing and things that we were working on that ultimately just felt like they would be easier to pursue as its own company. Now, five years later.
[0:02:28] The startup is valued at a reported $350 billion and named at Times 100 most influential companies of 2025. One of the big knocks from outside on artificial intelligence is this idea that...
[0:02:45] It's going to do the job for us. So we're not going to have to think as much as human beings. And we lose something when we allow the technology to do the thinking for us. Is there a way to build what you're building and still force humans to keep building our own intellects and our own free will?
[0:03:07] Don't necessarily want these models to be replacing the higher level thinking that they're doing. And we don't actually think that's the best use of the technology. We ran a campaign recently called Keep Thinking. And the idea behind this is that Claude is really a tool to help promote your ability to think more deeply, to solve problems, to be ambitious in the types of projects that you're taking on. And the sort of idea behind it is there's never been a worse time to be a problem. There's never been a better time to be a problem.
[0:03:36] And that's because I think if designed correctly, these tools can really enable people to get access to more information, right? You can actually help students learn better, learn, for example, to their own style of learning, as opposed to giving them an answer right out of the gate. You're a mom. I'm a mom. And our kids are growing up in a world where there will always be artificial intelligence. Does that excite you? Does that scare you?
[0:04:06] I think probably like most parents, I feel a mixture of things. I feel excitement about the opportunity that I think this will create. And I also feel some anxiety just about the world being different than when we grew up, right?
[0:04:20] For our parents, probably the advent of the internet was maybe the closest analog, right? What does it mean that your kids are able to go online and research anything or talk to strangers from around the world? And that was an entire sort of societal reckoning that we had to go through to figure out what that meant. And to me, I think AI is maybe analogous to that, but perhaps on an even bigger scale, right? I think there's some opportunities and some advances that we hope will come out of artificial intelligence that just make me
[0:04:48] Look at the situation with a huge amount of optimism, right? Something Dario, my brother, our CEO, has written about extensively is the potential for AI to really help with things like health care, looking for cures to diseases, vaccine development. I think there's a huge amount of potential. And when I look at my kids, I think, wow, it would be amazing if this technology just enabled them to live healthier, happier lives.
[0:05:12] I think on the other side, there's still a lot of work for us to do from a societal perspective to make sure that we're developing the technology thoughtfully, safely. How are we going to make sure that it's helping people live the lives they want to live, that it's not disempowering people? And how can we be transparent about the risks as we go about building the technology and really building the future? I'm curious about screen time. How much are you letting your kids use computers?
[0:05:38] Will you allow them to talk to AI chatbots, for example? So my kids are still a little young, so I feel like I get to have more thoughts about this as time goes on. But for my son, who's four and a half, I should start by saying Anthropic doesn't allow our chatbot to be used by kids under age 18.
[0:06:00] And the reason for that today is just that we're just not certain enough about what the impact is on children of artificial intelligence. It's possible it could be completely safe and completely fine, but the advent of this technology is so new that we would rather take a cautious tack.
[0:06:20] in almost every area that we can. That being said, I think there, of course, can be a place for talking to your kids about what AI is because it's going to be a part of their life, right? I think my son is right at around the age where he can kind of maybe understand, you know, what does mom do when she goes to work all day? And I think really explaining to him, hey, this is a really powerful tool, but I think our view today is we're just not sure enough about the impact that this is going to have on kids' brains that are still developing.
[0:06:49] That we would feel comfortable exposing it to them directly today. Yeah, I mean, I think that makes a lot of sense, especially as we see more and more of these reports of young people interacting with AI chatbots, leaning into delusional thinking, and it doesn't feel like the guardrails are there. And it's hard to even know if it would be possible to impose guardrails. What do you think about that?
[0:07:17] I think this is such an important topic, and it's one that I think, frankly, the industry as a whole has just not been talking about and grappling with enough, in my opinion. Our view on this is, first of all, like we talked about, kids' brains are still developing, right? And so the idea that kids under age 18 should maybe just have limits or parental controls on their abilities to sort of interact with the models.
[0:07:47] Directly feels like a good first step. Would you support a law? Anthropic has talked in a fair amount of detail with both California, at the California state level and New York state level, about the need for regulation, including on topics related to child safety. Obviously, we're excited about the potential for this at the federal level as well. 18 years old, you have to be 18 to use artificial intelligence or interact with a chat bot? I think it really...
[0:08:16] Depends on the situation is the way that I would maybe put it is you can imagine in a structured learning environment, right? Some of the companies that use Anthropics LLM that are educators have built in additional guardrails and have a human in the loop, an adult.
[0:08:32] Absolutely can imagine cases where that could be really beneficial for learning, right? But my feeling again today is we just haven't studied it that much. And so I think having a very cautious approach out of the gate is the right one. But additionally, I think there's work we can do at the model layer to help make these models more safe because
[0:08:49] Chances are kids might be exposed to LLMs regardless of how many safeguards are put in place, right? And so something we really work hard on is this topic of sycophancy, where sometimes these models will encourage sometimes maybe a person's in the middle of a mental health episode. It would be really bad.
[0:09:08] It's the models over encouraging that person sort of in a crisis state. And something we really have worked hard on is helping the models to not just say, yes, you're right, yes, you're right. Which by the way, becomes much harder if you're in an advertising based business, right? If you earn more money from having
[0:09:25] Your eyeballs, the customer's eyeballs on the model for a longer period of time, that's really not a great incentive for discouraging things like sycophantic behavior. Yeah, you want it to pat the user on the back so the user keeps coming back for more. Exactly. And this is really one of the inputs that we thought deeply about when making the decision to not do ads.
[0:09:45] We don't want an incentive system that basically says, hey, the more you talk to me, the more money my company earns. So let me keep talking to you regardless of what mental state you're in, right? And then again, this overall incentive structure, just like social media. I mean, anyone has sort of been on Instagram for too long and been like, I don't feel good, right? I've just been sort of clicking through this content of strangers. And I think AI, unfortunately, could just be another instantiation of that.
[0:10:12] But perhaps worse, right? Because it's a model that is designed specifically to get you to give it very personal, sensitive information to know a lot about you, the person. It's very hard, I think, for the person to not become the product. Which is essentially what social media has been. That's right. And I think we've really talked a fair amount about this idea that if you were a social media company 15 years ago and you could look into the future today...
[0:10:40] At that point in time, would you have felt good about what you saw? Would you say, wow, this is really the world that we wanted to build. We feel good about this outcome. There were a lot of small steps that happened along the way, many of which could have just been really reasonable decisions at each of those points in time.
[0:10:56] But I think what Anthropic has always tried to do is imagine ourselves 10 or 15 years in the future and said, okay, is there some way that we can try to predict what could some of the negative externalities of this technology be? And is there something that we could do to try and prevent some of those bad things from happening without, of course, being 100% able to see the future? So tell me, what does it look like 10, 15 years from now? So Dario has talked about this publicly a lot,
[0:11:26] Pretty soon, maybe less than 10 years from now, we could have something equivalent to a country of geniuses in a data center. A country of geniuses in a data center. That sounds lovely. Yes. So what does that really mean in reality? If you pull out your phone and you're able to ask Claude, for example,
[0:11:47] Hey, here are all the symptoms I'm having. Can you tell me what's making me sick? And Claude can answer that with as high level accuracy or maybe higher than a trained medical professional. What does that mean for society, for doctors, for patients? That's going to be a huge transformative event. And now you can imagine that for a lot of different disciplines. We're asking ourselves,
[0:12:12] How do we balance the incredible amount of good that could come out of this, right? We have this value called hold light and shade. And the reason we sort of invented this idea is we truly see both sides of this coin, right? We see the potential to cure disease, to help really increase access to information, level inequality.
[0:12:33] But on the other side of the coin is this idea that there are many risks, right? There's many things that could go wrong from how we develop AI, from what it means for society more generally. And we really try and talk as openly as we can, both about the good and the bad. Are you firmly set on the light outweighing the shade? So I'm an optimist, and I tend to believe that...
[0:12:59] The ability for us as a society to get through hard problems is more than we think. And I like to think that AI will just be the next instantiation of that.
[0:13:09] But I think there's a lot of risks as well. And I think our ability to properly mitigate the risks, to talk about them, to create thoughtful legislation, to partner together with policymakers and civil society, we're going to have to get all of those things right in order for us to realize all of the incredible benefits. And I think that is, that's a pretty tall order. It is. Yeah. And a lot of things have to go right.
[0:13:37] in order for this to not go really wrong. I think it's our job to help them go right. As the model providers, as the people that are developing the technology, I think Anthropic really views ourselves as trying to be responsible stewards of the technology. We can't just sort of say, well, we're going to create it and good luck.
[0:13:57] We have tried as much as possible to publish information and reports about what we're seeing. What are the actual impacts of AI on the labor force today? What are we really seeing in real time? Is it replacing jobs? Is it augmenting jobs? Is it a mixture of the two? And what we see is a mix. Some of the estimates are that AI could replace or automate 300 million jobs, 25% of the jobs in US and Europe. What do you say to the person who's
[0:14:26] Who's concerned by that? Who's afraid of that? The first thing I would say is we do not have all the answers. We don't have a perfect solution to potential automation of jobs, of some jobs. What I will say, perhaps again, overly optimistically, the percentage of jobs where you say an AI model can just do what this person does and their job is gone, that's a vanishingly small number today. It's not zero, right? Especially around things like customer support.
[0:14:54] But I think for even some of the most cognitively challenging tasks, the ability to augment with AI, I think will be very profound. I think this will be a shift. There are parts of it that may not always be graceful. And I think Anthropic is trying to kind of sound the alarm about that.
[0:15:13] And I continue to believe that humans plus AI together actually create more meaningful work, more challenging work, more interesting work, high productivity jobs, and that I think it will also open the aperture to a lot of access and opportunity for many people. What did you study in college? Literature. Literature. Would you study that again, knowing about AI? I think my answer on this is actually a little bit funny, which is that...
[0:15:42] I ultimately believe in a world where AI is very smart and capable of doing so many things, the things that make us human will become much more important instead of much less important. And what I mean by that is when we look to hire people at Anthropic today, we look for people who are great communicators, who have excellent EQ and people skills, who are kind and compassionate and curious and want to help other people because
[0:16:11] At the end of the day, people still really like interacting with people. And in an ideal world, AI will complement those skills, right? It will make those things about us better instead of worse.
[0:16:21] To go back to your direct question, I actually think studying the humanities is going to be more important than ever. A lot of these models are actually very good at STEM, right? But I think this kind of idea that there are things that make us uniquely human, understanding ourselves, understanding history, understanding what makes us tick, I think that will always be really, really important. And I think the ability to have critical thinking skills and learn how to interact with other people will be more important in the future rather than less. Coming up, what three years from now?
[0:16:51] What will Claude do that's gonna blow our mind? What's it like being at the forefront of all of this and working next to your brother? It feels just as crazy to us, I think, as it does to everybody else in the world. Working on it does not take away both the awe and the surprise of watching these models just be able to do more and more things.
[0:17:21] There's incredible potential, there's incredible risk, and we're really sort of trying to thread that needle. I feel very glad to be on this journey with Dario because, you know, we've always been close since we were little kids, and we always had this dream of getting to do something really big together. Who could I trust more in the world, right, than my sibling? What is it that makes Claude different from its competitors?
[0:17:45] One is this framework. It's called constitutional AI, but I think the best way to think about it is making the models helpful, honest, and harmless. I think the second one is, how do we actually ensure that we stay at the frontier of model intelligence? I think if there's one thing at the model layer, it's this desire to make sure that our models are always as capable as possible without ever compromising on safety.
[0:18:13] And is it possible, in your view, to make them helpful, honest, and harmless 100% of the time? There's actually some tension between those three areas. You might have a model that's perfectly helpful.
[0:18:27] But it might not always be 100% harmless. You might ask the model to do something that's kind of untoward or unpleasant, and the model actually has a challenge, which is, well, the user's asking me to do this, but it's kind of creating this sort of harmful output. What do I do? And so there's always a little bit of an inherent tension between those three that we think is healthy, right? And so much of what our teams, particularly our character team, works on is how do we develop these guardrails
[0:18:57] So harmlessness is not the priority. I don't think I would put it that way. I would say we have put in place fairly robust guardrails at a variety of different levels.
[0:19:26] The first is we've actually trained Claude with something called a constitution, where we've essentially said,
[0:19:34] Here is the system of ethics and rules that we would like Claude to follow in all situations, right? How do we actually give Claude a moral and ethical framework to understand what harmless really means? And clearly it's important to you, and it's something you talk about, it's something your brother talks about, but it's not clear that there's real progress being made out of Washington, for example. And even if there was progress made out of Washington, this is a worldwide phenomenon.
[0:20:03] This isn't just happening inside of our borders. The model providers alone cannot solve all of the challenges and all of the risks that are going to come out of AI. It is going to have to be a partnership with government at many layers, right? We've worked again closely with California state legislatures, more recently in New York. I think there's a lot of incredible progress that we are seeing happen. And I absolutely have faith that we will come to
[0:20:29] Thoughtful regulation that does not hamper the ability for this technology to continue to develop, right? It's of course going to be a balance, but my sense is that this is going to be a bigger and bigger issue for more people. People care about this issue. They want common sense regulation. And do you have confidence that governments are ready to do that? I think it is somewhat incumbent upon the model providers to view ourselves as functional experts.
[0:20:58] We have always tried to pioneer this idea that
[0:21:02] Anthropic, the company, actually owes something to regulators and to policymakers to help them make good decisions, right? The technology is developing so quickly. There's so much that's still unknown. And really, this partnership is just very important, right? We have the information about what the models can do. Sometimes we're the first to spot real world threats and risks because we just have all the data and the information. How do we arm policymakers, civil society, and other groups
[0:21:30] With the knowledge that they need to be able to make good decisions. Ten years from now, will there be all the AI companies that exist today? Who's going to win this race? Let me give you my best guess at where I think this might be going. I think we have long suspected that ultimately this is not going to be a winner take all.
[0:21:51] But that there will likely be a few major players that are sort of at the frontier. There's kind of Google, there's OpenAI, there's Anthropic. Then there's a few other model providers that are very strong but maybe one tier below. And then a long tail of other providers. It's possible I could be completely wrong and there's kind of one player that will take it all.
[0:22:10] But especially over the course of the past year or two, we've really seen more differentiation at the model layer, which is not something we necessarily expected, right? Anthropic is now known as pretty universally the best model for code, but we're not the best model for image generation because we chose not to do image generation for a lot of the safety reasons that we've talked about. To me, that level of differentiation
[0:22:34] Looks like we might be leaning towards a world of a few model providers that are sort of at the frontier. What three years from now will Claude do that's going to blow our mind? I mean, three years ago, everyone was astounded. What can these models do? You can talk to them. They can answer you. I couldn't believe it, what I was seeing. Yes, completely. Three years from now, I think the potential for these models to do things like...
[0:23:00] Help with drug discovery. Help us really understand so many of the health challenges that have faced humanity really for millennia. I think our potential to accelerate work there is almost limitless. We could be looking at a world of just
[0:23:16] Much more prosperity, much more opportunity, much more health. This is my dream. That's what I hope will be true in three years. But I think we have a lot of work to do to be able to get there. We really have to mitigate a lot of the risks that we've talked about effectively in order to realize this positive future.