Original video title: Inside the Mind of Anthropic CEO Dario Amodei | The Circuit | Extended Interview

Original source: Bloomberg Original
Original compilation: Web3 Sky City

Editor's note: Anthropic CEO Dario is currently in a very awkward position. On one hand, he holds the world's leading AI model; on the other, he was accidentally shut down globally due to a ban issued by the U.S. government, making even non-American team members unable to use it.

How this situation will end remains unknown. Rumor has it Dario is still working hard, so everyone can keep watching and commenting. But from this latest Emily interview, we can get a glimpse into the thinking of this controversial leader at the forefront of AI coding.

In today’s Silicon Valley power landscape, Anthropic occupies a highly unique and tense central position. As the strongest challenger to OpenAI, it was founded by top researchers who left due to ideological differences.

When CEO Dario sits under the spotlight discussing exponential growth in AI, he shows a rare, surgeon-like calmness. This is not just a technical competition; it’s a deep game involving trust, safety, and how human civilization should navigate the burst of intelligence.

Summary of the full interview

This latest interview delves into Dario’s mental journey amid AI’s exponential growth, covering behind-the-scenes stories of leaving OpenAI, the company's business model choices, and AI’s profound impacts on employment, cybersecurity, and geopolitics.

Dario elaborates on how Anthropic balances power through mechanisms like establishing a "Long-term Benefit Trust," and pursues technological leadership while setting "red lines" and delaying the release of high-risk models (such as Mythos) to uphold safety values.

Note: Dario’s remarks have often been unfriendly toward Eastern powers; please judge accordingly.

Core viewpoints

· The AI industry is experiencing a "steady exponential growth," which after accumulating enough quantitative change, will lead to a qualitative breakthrough.

· Trust is the cornerstone of AI industry cooperation. Anthropic advocates for trustworthy participants to unite and establish industry standards.

· Enterprise-level business models align better with AI safety values, avoiding addictive and low-quality content competition common in consumer markets.

· Society needs to anticipate and formulate macroeconomic policies for AI-induced unemployment risks, seeking a positive-sum game of "doing more with the same resources."

· Military applications must adhere to the principle of "human in the loop," strictly avoiding large-scale surveillance and fully autonomous weapons.

Below is the full interview:

Pressure and experience of exponential growth

Emily Chang: How much sleep do you get?

Dario Amodei: I’ve never been someone with very good sleep quality. I can only say I’m learning the art of relaxing and falling asleep under extraordinary pressure.

Emily Chang: Things are developing too fast. What does it feel like to be in the middle of it?

Dario Amodei: It’s a kind of exponential feeling. Like, suppose you’re on a spaceship accelerating at relativistic speeds away from Earth. According to special relativity, when you wake up from a nap, two days have passed on Earth. So you have to handle two days’ worth of things in one day.

Then you go back to sleep because you keep accelerating, and three days have passed on Earth. The next day, four days have gone by. That’s roughly what it feels like.

Emily Chang: Do you often feel anxious about what you might face upon waking?

Dario Amodei: We already have enough clear and urgent problems to handle. While dealing with these, I also think about how we should prepare. But I believe paranoia or worrying about what you’ll face upon waking is unhelpful. I’ve studied people who have handled high-pressure situations in history. You need to learn to respond rationally, not to compare the severity of various dangers.

This oscillation between "I don’t worry" and "Oh my God, we must panic today" I see as a sign of immature decision-making. Truly mature decisions are: you can’t ignore this, but you also can’t be complacent.

In fact, the risks are increasing, but we must respond rationally, like surgeons handling surgery, or military officers managing operations. Or, anyone making decisions affecting many people must do so rationally. They must understand the risks but also remain fundamentally calm.

So yesterday, my son asked if he could use my Claude account. I said absolutely not; I need my tokens. We see more of these applications in the consumer market. We initially wanted to be more of an enterprise company, but even without much effort, consumer applications are growing rapidly.

Emily Chang: You’re now at the center of the AI universe. How does that feel?

Dario Amodei: Interestingly, throughout my career, especially since joining Anthropic, I’ve experienced a kind of steady exponential growth.

This steady exponential growth feels like: nothing happens, nothing happens, then some small things happen, and suddenly, it explodes wildly. That’s the experience of this world. It’s also the experience of the company’s scale compared to others and the world.

So I looked at this chart for a long time, then I said, we’re probably at a point where we’ll become the highest-revenue and highest-valuation AI company. And indeed, that’s happened. It’s happened. So, in a way, I’m not surprised because the curve on the chart is very smooth.

But of course, from another perspective, when things really happen, you see more details and colors. And, that’s definitely surprising.

We always focus on questions like: how do we train excellent models? How do we apply them to great products? How do we ensure safety? How do we help people while managing the social risks around this technology? These are all the same questions, just viewed under a larger microscope.

Growth background and Silicon Valley spirit

Emily Chang: What kind of kid were you growing up in San Francisco? I know your father was a leather craftsman, and your mother worked in a library. How did that influence you?

Dario Amodei: Back then, the whole internet revolution was happening around me, but I had no interest in it. I was only interested in studying math and doing things like drawing and painting. I was very interested in exploring the universe. I loved science fiction. Overall, that was my environment at the time. I think I was just deeply curious about the world.

Emily Chang: You grew up in a place called the tech hub. Now it’s also the center of AI. Did anything in this place, this city, shape your worldview?

Dario Amodei: Yes. I think the non-conformist, individualistic, and “a little crazy is okay” spirit is very much present. A large part of it probably influenced me subconsciously.

You hear stories, for example, in some European countries or other regions of this country, where thinking differently is often suppressed or seen as strange, or where some crazy ideas are considered odd.

I have many criticisms of Silicon Valley, but I think one good thing is that it encourages a philosophy: even if all experts oppose you, it doesn’t matter. If you have a coherent vision and worldview, you should pursue it—that’s what matters. Maybe it’s fundamentally wrong.

But if it works, it has a certain long-tail effect. In some fields, you might dig deep and eventually find a huge gold mine. I think this spirit is very important.

Emily Chang: You, Daniela, your sister, and her husband Holden Karnovsky, all lived together in a shared apartment in 2016. What were you arguing about back then?

Dario Amodei: I think that was around the time Open Philanthropy was just starting. Holden was the head of that project. I was a biological scientist then, helping them with issues related to health in developing countries or biological research. So I was giving them some advice—like, which areas look promising, and which don’t.

Core reasons for leaving OpenAI

Emily Chang: Your decision to leave OpenAI has become a legend in Silicon Valley. What exactly happened? Setting aside the narratives, what’s the core issue? Where did you disagree?

Dario Amodei: Look, I need to say this very plainly. When you build powerful technology, you face many challenges, and Anthropic experiences these daily. We don’t always know whether the decisions we make are right or wrong.

So, there are reasonable disagreements about safety. We’ve had some differences with them, but know that that alone isn’t enough reason to leave. People here have disagreed with me. People here have disagreements among themselves.

But when you feel you can’t trust someone, when you feel their values aren’t what they claim, when you feel they’re not honest enough—when you see unsettling patterns or dishonesty, it becomes very hard to continue working at the company or trusting it. Ultimately, when you have no shared vision and no trust, why bother arguing anymore?

The solution is to go separate ways: you do your thing, they do theirs. I fully accept that: we operate in our way, they operate in theirs. We’ll see who wins in the market, who wins in the court of public opinion. I believe these facts speak louder than any dramatic speculation about who left and why.

We must remember that we are setting an example for how to deploy this technology responsibly. If they have objections, they should present arguments. I think there’s no need to say more about this.

Industry cooperation and rebuilding trust

Emily Chang: At the India AI summit, there was a moment when you and Sam Altman seemed to refuse to shake hands on stage. What happened?

Dario Amodei: The situation was extremely chaotic. We were only on stage at the last minute. They changed our positions temporarily. Then they took a photo of us, and ordered us all to hold hands. If you’ve attended such summits—I’m not criticizing India—they are very chaotic, especially those attended by heads of state.

Emily Chang: But everyone else held hands. Come on.

Dario Amodei: Look, I don’t know how to tell you this, okay? Narendra Modi was right there, suddenly calling everyone to hold hands.

Emily Chang: Okay, okay.

Narrator: See, Sam and Elon are suing each other. Looks like you don’t like Sam.

Emily Chang: If the world’s most important tech developers can’t even shake hands on stage, how can we trust you to cooperate on existential risks**?

Dario Amodei: That’s what I want to tell you. Among the people building this technology, there are huge differences in quality and trustworthiness. I think this means that different people believe no one trusts each other. I don’t think that’s right.

I know Demis Hassabis, who’s building the Gemini models, competitors to Claude. I’ve known them for 15 years. We’ve collaborated on many issues. We buy computing resources from Google. We often exchange safety ideas.

So, my view is that, first, some participants are more trustworthy than others. And I trust some participants outside of Anthropic. I think they are trustworthy. I believe what needs to happen is that trustworthy participants should unite to deal with untrustworthy ones. Forcing everyone to adopt the same standards.

Through extensive experience, I’ve learned that some people don’t do the right thing automatically. But if most people in the industry do the right thing, then others have no choice but to follow the trend. It’s a positive feedback loop—encouraging others.

It’s like Demis and I motivate each other. He’s working on AlphaFold. We’re trying to do some things in biology, like interpretability. They’ve started interpretability research. It’s not even really competition.

It’s just that each company is doing cool things. Others think it’s cool too. We want to try and see if we can make something new in this space. That’s the carrot in top-level competition. Then there’s the stick—or the hidden stick—where you realize these people are doing the right thing. If others don’t, it looks bad.

We often see behaviors like: they do the right thing grudgingly, but try to pretend they’re doing something different, hinting at some evil or malicious intent. That’s predictable. But I think that’s how we integrate the industry and promote cooperation.

Business models: enterprise alignment and values

Emily Chang: Earlier, others focused on interesting, flashy consumer applications. You bet on coding and enterprise, with Claude Code and Claude Co-work achieving great success. Why did you make that bet? Was it a value-driven decision or a business decision?

Dario Amodei: When we founded Anthropic, the most fundamental and always-important thing was our inner desire. We wanted to do this the right way. But you have to ask yourself: to fund these models’ extremely expensive development, the company must have a suitable business model. Does the business model hinder the realization of values?

This question always exists. But from my experience at other companies and observing others, I’ve learned that: if your chosen business model fundamentally conflicts with your values, you’ll be in trouble. You’ll either betray your values or be eliminated by the times.

You’ll end up in a dilemma; there are ways to avoid it, but it’s very tricky. Choosing a business model compatible with your values is much better.

So, when we think about this, we believe that, having seen the social media and consumer worlds, it seems to encourage interaction, even addiction. The kind of low-quality, mass-produced content we see in AI video models is driven by maximizing attention span, which is motivated by ad revenue.

But in the enterprise realm, our original intention was that these models should benefit people. When I think of all the positive things AI can do, I often remind people of the negative impacts. But fundamentally, we believe the positive side will ultimately outweigh the negative. Many positive applications fall into the enterprise category.

We want to use AI to cure diseases that were previously incurable, which requires collaboration with biotech and pharmaceutical companies, as well as academic research. All of these are enterprises. We want to use AI to make energy cheaper and more efficient. These are all enterprise applications.

We want to use AI to improve education. Most of these are enterprise-level. We want to use AI to solve health and development issues in developing countries. Though they are non-profits, they are essentially enterprises. We want to promote economic growth. That’s also fundamentally enterprise-level.

Additionally, I think another factor is that enterprises value trust and long-term relationships. Consumer applications sometimes have a gimmicky feel, but in enterprise, it’s about building partnerships—working with a company for years, keeping promises, and earning genuine trust.

Therefore, this aligns very well with our goal of deploying models responsibly and safely. I believe having a business model that closely matches our values is very advantageous.

It’s not that conflicts never happen, or that we don’t have to make tough choices. But I think such choices are far fewer than in other situations.

Competitive moats and the "SaaS apocalypse"

Emily Chang: Developers can switch from Claude to GPT or Gemini in a single afternoon. Is it really possible to maintain long-term leadership in this industry? And how long do you think a serious competitor would take to replicate what you’ve built?

Dario Amodei: Model quality is the most important thing. For example, we are currently far ahead in model quality. Although there’s some inertia, I’ve never relied on it—I’ve never depended on the idea of “high product stickiness, users won’t switch.”

I think you still want a better model. You want a better product. And we see no slowdown in growth rate; if anything, it’s accelerating, at least as of this interview. So I believe that’s the most important thing.

Emily Chang: Shortly after Claude Co-Work launched, the market value of $285 billion evaporated overnight, and traders called it the SaaSpocalypse. If AI continues to improve at this pace, how much of traditional software will be replaced, and how fast?

Dario Amodei: That’s a very hard question to predict in advance. If you could predict perfectly, people would have already done so, making huge profits and always being right.

So, no one knows exactly what will happen. But I’ll point out a few things: all traditional software companies have some moats. I think some of those moats will disappear, but others will remain. The ability to rapidly develop software—I believe that will vanish. If your moat is “we’ve built complex software no one else can,” good luck. You won’t be able to defend that.

But I believe that relationships with customers, domain expertise, and specialized knowledge are valuable. So my advice to all these people is: don’t become complacent. Don’t ignore it. List all your moats, and be very clear that some will disappear, while others will become more important because they are limiting factors. New moats may also emerge.

I think those who can adapt skillfully—leveraging existing moats and creating new ones—will do very well. I believe those who are complacent, who think that old methods will always work, will have a hard time. So that’s my advice.

And I think, ultimately—I guess, depending on how you define SaaS and what doesn’t count as SaaS—the software industry will grow larger, not smaller, even though some big players will fail.

Emily Chang: Please explain.

Dario Amodei: I just think the overall size of the cake is getting bigger. For example, with AI, that cake is expanding. Established companies might shrink relatively. Some may lose value if they don’t adapt properly. Some might even go bankrupt.

But I believe that when growth is very rapid, it’s common to see this: if AI’s potential increases tenfold, existing industries might grow 1.5 times—just not as much as the whole cake. So I think this scenario is likely.

That doesn’t mean we won’t see some huge losers. I think those who fail to adapt, who bury their heads in the sand, who can’t see future trends, or can’t recognize their own moats—will face very tough times.

Compute power, funding, and partners

Emily Chang: Your biggest supporters are companies like Amazon, Google, Microsoft, and NVIDIA. These companies have their own agendas—they’re both partners and competitors. You have huge milestones tied to funding. Who’s really in charge?

Dario Amodei: We’ve been very frank about our views. I’ve always been very clear about supporting export controls on chips to China. I say this because I believe that if China leads in AI capabilities, it would be very bad for the U.S. and the world. Some chip manufacturers obviously disagree. But that hasn’t stopped me from expressing my opinion.

Even after signing more cooperation agreements, I still reiterate this. They understand that we’ve always cooperated with them. We’ve been good partners. We can work together. I’m sure they wish we wouldn’t say these things, but I firmly believe in what I say.

What will I do? Ultimately, they will reach an agreement. They benefit from these deals just as we do. Look, we’re all adults. We can cooperate on some issues while disagreeing on others.

Emily Chang: Bloomberg reports that your valuation even exceeds OpenAI’s. We’re talking about a startup founded five years ago, valued close to a trillion dollars. How do you interpret that? Regarding that number, and since you’re more disciplined in compute resource management and have a faster path to profitability, why do you still need so much money?

Dario Amodei: The pace of compute expansion is very fast, so the business fundamentals look good. But in a year, your compute scale might be three or four times what it is now—I won’t give exact figures, but this kind of compute growth is very rapid.

We have every reason to believe that revenue will grow to and beyond that scale. Raising funds is a buffer against this uncertainty.

So, it’s a completely rational approach. The dilution of equity caused by this is very small; logically, they are not the same thing. In fact, it’s compatible with the opposite—this doesn’t mean the fundamentals are bad.

Emily Chang: There are reports of server overloads, reliability issues, and complaints about tokens running out. You said other companies are going all-in on infrastructure. Do you really have what you need, or are you just chasing the pace?

Dario Amodei: Regarding compute, one point is the so-called marketing compute. My view is that, over time, even over several months, we can acquire large amounts of compute. It’s worth noting that, I don’t think we’re buying too little compute by any reasonable standard.

Our plan was to grow compute tenfold each year.

Tenfold annually is what we expected. But that’s not what we’re seeing now. In Q1 2026, our quarterly revenue grew more than three times, just for one quarter—not annualized. Growing three times in a quarter, of course, means an annualized rate of 80 times.

We didn’t expect an 80-fold annual growth. Planning for 80 times annual growth is irrational because if you only achieve ten times, you’re missing out on eight times. So, we’re in a phase of explosive compute growth. This can’t last forever.

If it continues, by the end of the year, you’d have a revenue figure no other company could match. I don’t think that will happen. It’s just impossible to sustain.

But you might go through a brief period where you’re amazed—wow, this growth is faster than anything we’ve imagined. But I don’t know—you’ve seen the compute agreements with Google and Amazon. We can and will do more.

For example, the market is fluid. If you can use compute very efficiently and there’s demand, you can get the compute you need. It might just take a month or two.

Leading the race and company culture

Emily Chang: How does it feel to surpass your rivals**?

Dario Amodei: Look, we still face many tough challenges. We have a “race to the top” mentality, trying to push other companies to keep up with us. I think we’ve already seen that we’ve driven them. Sometimes they don’t admit it.

Sometimes they attack us while copying us, but this kind of motivation is very valuable.

So I think, from a business and model perspective, being an industry leader isn’t about beating competitors. It’s about leading the entire ecosystem to develop together. We want to do more in this regard in the future.

Emily Chang: But I have to say, winning does feel pretty good.

Dario Amodei: You see, we always strive for success, right? We’ve been working hard. We’re not trying to fail here. I’m not the type to think we should stop developing this technology or building it. We operate within a free enterprise system, and that’s fine. We just need to mitigate the risks of these models, so it’s always about balancing the two.

Emily Chang: So, for most of Anthropic’s history, you’ve been at a disadvantage. I think when you have nothing, it’s easier to hold the moral high ground. How hard is it to stay true to your principles at this scale?

Dario Amodei: I think about this a lot.

As the company grows, I stay vigilant at every stage. Every phase of growth brings new challenges. The company might lose its character in some new way—whether it’s commercial ambition or core values.

I worry about both because I think they are synergistic.

In fact, I believe the reason we can build such excellent models is precisely because it allows us to practice our values effectively. As the company grows, there are many pitfalls. There are many ways things can go wrong—not because my values, or those of the co-founders or leadership, have changed, but because the company’s composition is changing very fast.

So I spend about half my time discussing Anthropic’s culture and how it operates, especially when you’re hiring many people from big tech.

If you don’t tell them how Anthropic works, they’ll just repeat what they’re most familiar with—the way their previous companies operated.

So, it’s an ongoing struggle and challenge.

It’s like, Daniela and I’s most important task is figuring out how to keep this culture alive. Because we realize that, in the long run, that’s what matters most.

Research efficiency and scientific progress

Emily Chang: Your product iteration speed is astonishing. The number and speed of your releases are incredible. How do you do it?

Dario Amodei: I’d say two things. First, we are a unified company. We have a unified corporate culture. I think, as we scale, we’ve maintained very high efficiency. Everyone shares the same goals, which reflects cultural and organizational unity. I think that’s the most important factor.

Second, I’d say it’s Claude itself. We’re now using Claude to help us develop models, improve efficiency, and rapidly build products. To do this, you need to develop various new practices. Though we’re still newcomers in this area, it’s already brought significant acceleration, and that acceleration is becoming more reliable. Those are the two main factors I’d point to.

Emily Chang: Can you tell me about the craziest thing AI has done that you’ve seen?

Dario Amodei: I think the craziest things I’ve seen mainly happen in biology and medicine. I’ve seen many cases, including Daniela’s, where Claude diagnosed some medical issues that many top doctors missed. In biology, these models are starting to perform astonishingly well in drug design, computational chemistry, and related tasks. As a former biologist, I can’t help but marvel at how difficult this is.

To do these tasks requires extensive specialized training. And Claude is becoming increasingly good at it. That’s a field where I believe we’ll gain huge benefits. This is the positive side—we’ll gain enormous welfare. Life will get better. The quality of human experience will improve.

Emily Chang: A century of scientific progress.

Dario Amodei: A century of scientific progress, and a century of improvement in human experience. For example, looking back to 1900. Think about all the problems we faced then, all the causes of premature death, all the suffering people endured, and all the material shortages we no longer face today.

Then imagine the next hundred years. I truly believe that, with scientific and medical advances, if we can overcome current challenges—I believe we can. I’m increasingly optimistic. We will have a much better world than today.

Writing, thinking, and AI assistance

Emily Chang: I know you love writing. You’re known for your articles. Do you use Claude to assist with writing?

Dario Amodei: I do. I haven’t yet let Claude write entire texts for me because I have my own style and am quite picky about it. But I mostly use Claude to help brainstorm, organize themes, or provide reference content.

It’s a supportive role. I don’t know how long it will take before Claude can produce work better than mine. We’re not there yet, but I believe that day will definitely come.

Emily Chang: I also love writing. I think writing helps clarify your thoughts. It involves a lot of critical thinking. If we let Claude do the work, will we lose that ability?

Dario Amodei: I’m a bit worried about that, and that’s partly why I insist on writing myself. It’s true for external audiences. Even though many read my articles, writing also helps me clarify my own thinking, figure out what to do next, and establish a shared reference point with others.

I think we’re still exploring how to use AI in a way that preserves these benefits. I think what I’m doing now—using Claude for research and to help organize my thoughts—is a good example.

I believe if we just rely on it entirely—like having it write an article about AI risks—first, it can’t capture my personal insights, and second, I would lose the benefits of writing itself. As models improve, I think there will be ways to incorporate them more directly into writing while retaining those benefits. But it will be a delicate process. The situation won’t be straightforward; we’ll need to explore gradually over time.

I think we might see a very counterintuitive combination: rapid GDP growth alongside high unemployment, or at least underemployment, with many low-wage jobs and severe inequality.

Unemployment risks, productivity, and macro policies

Emily Chang: You’ve been very outspoken about unemployment. AI might eliminate half of entry-level white-collar jobs in 1 to 5 years. That was last year’s statement. AI’s pace is astonishing. Is it still 50%, or has it increased?

Dario Amodei: I’ve always said, if you look back at those original clips, they’re often taken out of context in those three seconds. But my real statement has always been: I don’t know what will happen, but it indicates that the situation could produce a shock of an order of magnitude.

And I’ve also talked about what measures we can take—I’ve mentioned token taxes and working with companies to adjust staffing. I’m somewhat skeptical of retraining programs, but we should definitely consider them in macroeconomic policies.

From the dawn of technology, I’ve discussed solutions in about five pages, explaining the difference between tasks and jobs, why this time is different, and listing six things we can do—from private charity to government action. I’ve talked about the problems and the solutions.

But social media, I despise it, as a category. People only clip three seconds from a year ago, never reading the full article, or exploiting social media’s features—I’ve already discussed these risks very carefully in my writings.

The idea that it’s just cheap marketing is itself a form of cheap marketing. It’s laziness. It’s failing to engage with serious intellectual work. I think that’s part of the problem. Again, I see this as a stubborn disease of Silicon Valley. It’s trapped in this three-second social media world.

So, people only react to this content, or think they only need to react. Again, I think that’s very dangerous, and we’ve failed to have a mature dialogue.

Instead, people lazily respond to these three seconds. Then they say, Oh, that’s what Dario said. That’s so stupid. That’s so unserious. Every time someone says that, my opinion of them drops.

National security and military AI red lines

Narrator: One of the world’s leading AI companies is deeply embedded in U.S. national security, including military operations. Anthropic’s confrontation with the Pentagon over AI military safety measures is intensifying.

Emily Chang: You’ve long held an anti-war stance, dating back to your Caltech days. Yet, you’re one of the first AI companies to sign contracts with the Department of Defense, operating on U.S. classified networks for warfare. Explain.

Dario Amodei: Okay. I want to say that, you see, the world is constantly changing. As I see this technology, I worry because we’re facing a resurgent authoritarian bloc that’s very aggressive, and we need to defend ourselves.

That’s been my firm belief all along, and I still hold it. That’s why, during two administrations, I may not agree with every policy, but I generally support this stance.

That’s why we cooperate with them. Of course, it’s not for money. It’s very troublesome—deploying on government networks doesn’t pay much but consumes a lot of effort. So, we do it out of a sense of mission. But since we’re engaged because of concern, the use of this technology must be limited.

[When discussing technological development stages, I’ve said: We should leverage this technology in all aspects, except in ways that violate our values. The red lines of mass surveillance and fully autonomous weapons are exactly what I believe would violate our values.

If these methods are adopted, even winning would be meaningless. That’s the balance point I see, and our position. It also explains why we’re among the first to cooperate with the Department of War, while choosing not to engage in certain activities others are willing to pursue.

I think you need to choose a stance and stick to it. The wavering approach—initially claiming not to cooperate with the government, then suddenly accepting all government demands—I just can’t understand that. You should pick your principles and uphold them.

Emily Chang: You’ve been working with Palantir since 2024.

Dario Amodei: Yes.

Emily Chang: Their technology has been used by ICE and police in Gaza. Is Claude used in other ways for surveillance?

Dario Amodei: We have not cooperated with ICE through Palantir or any other channels. We do not work with CBP. I don’t think we have business in Gaza. We are very cautious about limiting our scope of cooperation to what we agree with.

Emily Chang: So, you’ve set your red lines—your president forbids cooperation with the federal government, the Pentagon marks you as a supply chain risk, and OpenAI has signed contracts you refused. What does winning this fight look like?

Dario Amodei: I don’t think a private company can win this fight. It’s not really a fight. Anthropic isn’t trying to win or lose. It’s more about a debate on how the government should properly use AI. AI is an emerging technology. We don’t yet understand where it’s reliable, where it’s not, how it can promote our values, or where it might undermine them.

So, one of the important things is to establish precedents for the good use cases—frankly, most use cases are good—and for the ones we worry about. As I said, we’ve seen that contracts have limited power. Others might sign contracts that don’t respect your bottom line.

But they raise awareness. Currently, Congress is working on serious bipartisan efforts to ban some of the issues we’re concerned about and to set safeguards. Reiterating, I don’t see this as a fight, but to some extent, it’s helping our country think more carefully about how to use this technology properly.

Values and military ethics

Emily Chang: Anthropic’s operators are ideological fanatics who shouldn’t have sole decision-making power. I care about broader AI issues. Do you mind being called an ideological fanatic or a bunch of left-wing lunatics?

Dario Amodei: I’ve been called much worse. People can call me or Anthropic whatever they want. The two most important things are: as a company, we’ve succeeded, and we’ve stuck to our values. In fact, in some ways, my life is very easy because when you only pursue those two things, it’s really simple, isn’t it? Like that, you always know your stance.

Emily Chang: An American official said that, with the help of LLMs, the U.S. military has increased its daily strike targets from 1,000 to 5,000. That means Claude can help kill more people faster. Are you comfortable with that?

Dario Amodei: I think there are two things. First, the U.S. military’s ability to increase efficiency. I support that ability. I believe having a stronger capability doesn’t trigger war; it acts as a deterrent. Basically, you’re asking whether I believe in this country, and whether I want this country to be a stronger actor rather than a weaker one on the world stage. I do. I am a patriot.

Second, a separate issue: are there policies the U.S. government is involved in that I might support or oppose?

The government’s involvement—obvious. I support some, oppose others. It’s not up to me. If we provide a technology, and the DOD raises this point, and we agree with their view, we have no say over what military actions they take or don’t take.

Right now, I might think some military actions are reasonable, others are bad ideas, but we won’t deny the role of this technology.

You must leave policy-making to military decision-makers. All you can do is set some high-level boundaries—prevent uses that conflict with our values or our country’s values, and promote uses that align with our values. That’s our approach.

Emily Chang: Bloomberg reported that the U.S. military used Claude in the Iran war, via the Maven Smart System platform built with Palantir, for AI-assisted target identification. It was reported that in February, a U.S. missile hit a girls’ school in Iran, killing over 150 people, mostly children. Did Claude play a role in that attack?

Dario Amodei: We, look, we don’t have access, and we don’t know exactly how these models are used. Obviously, mistakes in war are very, very terrible, and that’s a very frightening thing.

If that can’t explain why we oppose those uses we don’t support, we’re even willing to risk damaging our company’s reputation to restrict how these models are used. And