Source: Bloomberg
Compiled by: Felix, PANews
Bloomberg recently conducted an interview with Anthropic CEO Dario Amodei, discussing his entrepreneurial journey in San Francisco, competition with OpenAI, and the ultimate goals of AI.
PANews has compiled the highlights of the interview.
Host: You are currently at the center of the AI universe. How does that feel?
Dario:Throughout my career, especially during my time at Anthropic, what I've experienced is a smooth exponential curve. The feeling of this smooth exponential curve is: nothing happens, nothing happens, a small thing happens, and then 'whoosh'—explosive growth. This is the world's experience, and also the company's scale experience relative to other companies and the world. I looked at this chart for a long time, I used to say 'we'll probably become the AI company with the highest revenue and valuation around this time,' and it indeed happened. So on one hand, I'm not surprised because it's just a smooth line on the chart; but on the other hand, when things actually happen, you see more details and colors, which is absolutely surprising. We always need to keep in mind the most fundamental questions: How to train good models? How to put them into good products? How to ensure safety? How to help people while managing the social risks brought by technology? They are all the same questions, just examined under a larger microscope.
Host: You grew up in San Francisco, what were you like as a child? I know your father was a leather craftsman and your mother worked in a library. How did that influence you?
Dario:The first internet revolution was happening around me at the time, but I had zero interest in it. I was only interested in doing math problems, doodling, understanding the universe, and science fiction. That was my overall environment back then, and I was filled with curiosity about the world.
Host: You grew up in this hub city of technology, and now it's also the center of AI. Has anything about this city influenced your worldview?
Dario:The general spirit of non-conformity, individualism, and 'it's okay to be crazy' here has indeed influenced me. If you go to Europe or even other parts of this country, you'll find that thinking differently or having crazy ideas is not encouraged and is even considered strange. Although I have many criticisms of Silicon Valley, I think one thing it does well is: It's okay even if all the experts are against you. If you have a coherent vision for the world, you should pursue it. Maybe it completely doesn't work, but if it succeeds, it creates a long-tail effect, and you might find huge gold veins in certain ore deposits. I think this spirit is very important.
Host: Your decision to leave OpenAI has become Silicon Valley lore. What actually happened? Besides the rumors, what were the real issues? Where did the disagreements lie?
Dario:Let me put it simply. When developing powerful technology, you face many difficult problems, and Anthropic also faces these dilemmas daily where we don't know if our decisions are right or wrong. So there are many reasonable disagreements on safety issues, and we did have some of those with them, but that alone wasn't enough reason to leave. Even within Anthropic, people have disagreements with me and with each other. However, when you feel you can't trust someone, when you feel their values are not as they claim, when you feel they are dishonest, not acting from the motives they profess, when you see disturbing patterns of deceptive behavior, it's hard to continue working with and trusting that company. Ultimately, if you and others have different visions and don't trust each other, why argue? The solution is to go your separate ways. I am perfectly at peace with us doing things our way, and them doing things theirs. See who wins in the market, who wins in public opinion. That is more telling than any dramatic gossip about who left whom. We are demonstrating an example of how to deploy this technology responsibly. If they disagree, they can present their arguments. I think that's all there is to say.
Host: At the AI summit in India, you and Sam Altman seemed to refuse to hold hands on stage. What happened there?
Dario:The truth is that summit was extremely disorganized. We all went on stage at the last minute, they changed our order on the spot, took a photo, and then suddenly ordered all of us to hold hands. If you've been to these international summits with heads of state, you know they tend to be super chaotic (and I'm not singling out India).
Host: But everyone else held hands.
Dario:I don't know how to explain it to you. Narendra Modi just suddenly called for everyone to hold hands on stage.
Host: Okay. Sam and Elon are suing each other. You don't like Sam. If the people building the world's most important technology can't hold hands on stage, how can we trust you'll collaborate on risks that threaten human existence?
Dario:What I want to tell you is that there are huge differences in the quality and trustworthiness of the people building this technology. That 'nobody trusts anybody' narrative is not correct. I've known Demis Hassabis, who developed the Gemini models, for 15 years; they are competitors to the Claude models, but we've collaborated on many issues. We buy compute from Google, and we often exchange ideas about safety. So I think, first, some actors are indeed more trustworthy than others; second, there are actors outside Anthropic that I trust. What I think needs to happen is for the trustworthy actors to band together, putting the untrustworthy actors in a position where they have no choice but to adopt the same standards. In my experience, some people won't do the right thing on their own, but if most people in the industry are doing the right thing, others don't have much choice. This has two aspects: The positive side is mutual inspiration (i.e., racing to the top), like Demis doing AlphaFold inspired us, we do biological research, and our interpretability research inspires them—it's not zero-sum competition. This is the positive 'carrot' side of the race; and the 'stick' side is that you'll look terrible if you don't do the right thing. They might reluctantly do the right thing while pretending to do something else, even thinking we have some sinister hidden agenda—that's to be expected, but that's precisely how industries align and cooperate.
Host: Early on, others focused on fun, flashy consumer applications, while you bet on coding and the enterprise side. Claude Code and Claude Cowork have been very successful. Why did you make that bet? Was it a values decision or a business decision?
Dario:When we founded Anthropic, the most basic thing was we wanted to do things the right way. But to fund the extremely expensive model creation, it needed a business model. If the business model fundamentally conflicts with your values, you're in a bind: either betray your values or become irrelevant. It's much better to choose a business model compatible with your values. We looked at social media and the consumer world, which often incentivize user engagement and even addiction (like the AI video model mess we see), just to maximize user screen time for ad revenue. In contrast, on the enterprise side, we hope to use AI to cure previously incurable diseases (collaborating with biotech, pharma, and academia), make energy cheaper and more efficient, help education and nonprofits, and promote economic growth. These are inherently enterprise applications. Another factor is that enterprises highly value trust and long-term relationships. The consumer side can sometimes be gimmicky, while the enterprise side requires you to build trust over many years and deliver what you promise. This is very synergistic with our goal of deploying models in a positive, safe manner.
Host: A developer can switch from Claude to ChatGPT or Gemini in an afternoon. Is it really possible to maintain a long-term lead in this industry? How long would it take for a strong competitor to replicate what you've built?
Dario:Model quality is the most important thing. We are far ahead in model quality right now. While there is some switching inertia, I (and Anthropic) have never relied on product 'stickiness' to keep people from switching. You need to have a better model, a better product. We see the growth rate hasn't even turned downwards. At least at the time of this interview recording, if anything, it's still rising. So I tend to think model quality is the most important thing.
Host: Soon after the launch of Claude Cowork, $285 billion in market cap evaporated overnight, with traders calling it 'SaaSpocalypse.' If AI continues to advance at this pace, how much traditional software will be replaced, and how fast?
Dario: It's hard to predict this accurately ahead of time. All traditional software companies have their own moats. Some of those moats will disappear, but others will remain. 'Quickly writing complex software' as a moat will definitely disappear; you can no longer rely on that for defense. But customer relationships, knowledge of how industries work, and unique domain expertise will still matter. My advice to these companies: Don't be complacent. List all your moats, see which ones will disappear, which ones will strengthen, because new moats will also emerge. Those who are flexible will do well, and those who deceive themselves and rest on past laurels will have a hard time. Overall, I think the software industry will get bigger, not smaller, though there will certainly be some big losers.
Host: Explain that.
Dario:Because the 'pie' is getting bigger. If AI raises the industry ceiling by 10x, existing businesses could easily grow 1.5x, even if their relative share of the total pie becomes smaller. But for those who don't adapt, who bury their heads in the sand, they will face tough times or even go out of business.
Host: Your biggest backers are companies like Amazon, Google, Microsoft, and Nvidia. They all have their own agendas, being both partners and competitors. Your huge business milestones are tied to funding rounds. Who is really in charge?
Dario:In many cases, we've been very upfront about what we truly think. We're all adults; we can agree to disagree, collaborate on one thing while disagreeing on another.
Host: There are reports that your valuation is higher than OpenAI's. For a five-year-old startup, a valuation nearing a trillion dollars. How do you make sense of that number? If you're more disciplined on compute and have a faster path to profitability, why do you need that much money?
Dario:The demand for compute is growing incredibly fast. Business fundamentals are good, but you might need three or four times more compute within a year; growth is extremely fast. We fully expect revenue growth to match or even exceed that compute growth. Raising money is to buffer the impact of that uncertainty; it's very rational and results in very little dilution of company equity. It absolutely does not indicate problems with the business fundamentals.
Host: Does it feel good to surpass your main competitor?
Dario:We have many difficult challenges ahead. We are promoting the 'race to the top' idea, trying to pull other companies forward alongside us. Sometimes they don't acknowledge it, sometimes they attack us while copying us, but that pulling effect is incredibly valuable. The value of being the leading company is not merely about beating opponents, but about having the ability to drive the entire ecosystem.
Host: But winning must feel a bit better.
Dario:We certainly always strive to succeed and don't want to fail. I don't think we should shut down this technology; we compete in a free enterprise system and just need to mitigate model risks. We are always seeking balance between the two poles.
Host: Your product development speed is astonishing. How do you do it?
Dario:Two reasons: First, we have a unified culture and organizational efficiency; everyone is aligned. Second, it's Claude itself. We are now using Claude to help develop our models and products, making the process more efficient. This generates a massive and increasingly reliable acceleration effect.
Host: What's the craziest thing you've seen AI do?
Dario:It's all about biology and medicine. I've seen several instances where Claude diagnosed medical issues that senior doctors missed. In biology, models are becoming astonishingly good at drug design or computational chemistry. As a former biologist, I know that requires an extremely high level of specialized training. This is the huge positive side of AI; it will greatly enhance human quality of life. Imagine the scientific progress from 1900 to now over a century, and then imagine if we had another century of progress. If we navigate the current challenges successfully, we'll have a much better world.
Host: I know you love writing. Do you use Claude to help you write?
Dario:Yes, but I haven't had it write full texts for me yet because I'm a bit particular about specific styles. I mainly use Claude for brainstorming, organizing themes, or providing references. It currently plays a supporting role. While it's still some distance from writing better than me, that will come sooner or later.
Host: Writing helps clarify thoughts, involving a lot of critical thinking. If we let Claude do it, do we lose that thinking ability?
Dario:I am somewhat concerned about that, indeed. I insist on writing myself, half for the reader, but equally to clarify my own thoughts. If you just tell it 'write an article about AI risks,' not only will it not produce what I truly want to express, but I also lose that benefit of thinking. In the future, as models get better, we may need to find subtle ways of using them more directly while still retaining the benefits of thinking.
Host: You've been outspoken about job loss, predicting AI might eliminate half of junior white-collar jobs within 1 to 5 years. That was a year ago. Is that proportion 50% now, or higher?
Dario:If you look at the original full video, my actual statement was: I don't know what will happen, but this represents an order of magnitude for how crazy things could get. However, psychology always leans towards clipping out those three seconds of 'doom is coming,' while ignoring all the solutions I simultaneously discussed (like token tax, corporate adjustments, macroeconomic policies, etc.). My message was never 'doom is coming,' but that we need to see it coming, focus on it, and proactively manage it. In the short term, AI will make people more efficient, but this is just a transitional phase. As we've seen in the history of automation, AI may eventually take over those tasks. Take Anthropic's software engineers as an example: right now, AI helps them write most code, making them more efficient, but we're also starting to see cases where it's simply better to let AI directly perform specific tasks. On the other hand, this also creates new demands, like 'applied AI solution architects (frontline deployment engineers)' who combine technical and customer communication skills. Job destruction and job adjustment will happen simultaneously.
Host: You released a chart showing potential job disruption, which ones will disappear, which will be created?
Dario:It's hard to predict precisely, like predicting the stock market. But overall, junior white-collar positions (in banking, finance, etc.) are highly likely to be replaced by AI. In our conversations with enterprise customers, we see them facing a choice: lay off to cut costs, or use the same resources to do more new things? We always try to push them towards a 'positive-sum game': use efficiency gains to do more, not to lay people off. As the 'pie' expands, people might find new places to go; it's a question of matching speed. The scale of disruption will be large, which is why I tried to warn people.
Host: If there's unemployment at that high a scale, wouldn't that be a cause for revolution?
Dario: That is absolutely an outcome we want to avoid. I think several areas will retain opportunities: one is the physical world (because robotics development lags behind AI information processing speed; manufacturing and construction will still require a lot of human labor); another is all human-centric work. Even if AI is more accurate than doctors or better at customer service, people still want and need to talk to other humans and build connections on important matters. Humans will also retain the role of guiding AI values and intent.
Host: Many people have pushed back on your views, like Jensen Huang saying you confuse 'tasks' with 'jobs,' and others calling it 'doom marketing' that benefits Anthropic.
Dario:I want to rebut this clearly. In our reports and papers, we spent five whole pages detailing the distinction between tasks and jobs, explaining why this time is different, and proposed six solutions covering both private and government sectors. But I've found a sickness in Silicon Valley and social media is laziness; people just watch a three-second clip from a year ago and irresponsibly comment. Calling serious intellectual work 'cheap marketing' is itself a cheap marketing tactic and an incredibly lazy refusal to engage in serious discussion.
Host: As a globally leading AI company, you are deeply involved in many areas of US national security. You've had a long-standing anti-war stance since Caltech, but you were among the first AI companies to sign a classified cyber operations contract with the Department of Defense. Please explain.
Dario:The reason is the world is changing. When I see Russia invading Ukraine, I worry about the resurgence of authoritarian blocs. We need to defend ourselves. We absolutely do not want adversaries using AI to analyze intelligence, attack Ukraine, while we are unable to defend. So we are providing support across government departments. By the way, we do this absolutely not for money; running on government networks is actually a big hassle and yields little profit. Precisely because we do it out of 'caring,' there must be clear limits and red lines on the use of the technology, namely: no mass surveillance and no development of fully autonomous weapons. If we abandon our democratic values just to win a competition, that's not worth it. This is the balance and stance we uphold.
Host: You started partnering with Palantir in 2024; their tech is used by US Immigration and Customs Enforcement (ICE), police departments, and even in Gaza. Is Claude being used for surveillance?
Dario:We do not work with ICE and CBP, either directly or through Palantir. To my knowledge, we also do not operate in Gaza. We are very careful to limit our involvement to areas aligned with our values.
Host: So you drew red lines, the President banned some of your federal authorizations, the Pentagon labeled you a supply chain risk, and OpenAI stepped in and signed contracts you wouldn't. What does winning that fight actually look like?
Dario:I don't think of it as a fight to 'win or lose'; it's a societal debate about how governments should properly use AI. We are trying to establish usage precedents concerning worrisome scenarios and have drawn attention to the issue. We see bipartisan efforts in Congress attempting to ban high-risk applications and set up guardrails. If there's a win, it's getting our nation to think more carefully about the appropriate boundaries for using new technology; that's the victory.
Host: Some describe you as an ideological lunatic who shouldn't have this much power, or a left-wing zealot. Does that bother you?
Dario:I get called names often, and much worse than that. People can call me and Anthropic whatever they want. I only care about two things: the company succeeding, and us sticking to our values. It actually makes my life very simple; I've always known where I stand.
Host: Is an AI war more likely to prevent World War III or more likely to cause it?
Dario: Overall, more likely to prevent it. But without usage restrictions, it could cause one. Without proper regulation, both sides could easily erupt into conflict due to misunderstanding. If we have overwhelming superiority in intelligence gathering and response capability, adversaries will think twice before acting. Superior intelligence can effectively deter conflict.
Host: Anthropic is in the headlines almost weekly, recently about the Mythos model. You said Mythos is too powerful to release to the public. What surprised you most about it?
Dario:What surprised me most is that it not only finds vulnerabilities but can also turn them into substantive attack tools; there's been a huge leap in that capability. Some early companies testing it even asked us not to release it, calling it a 'super weapon' that should require a gun license to use. Our decision not to release it publicly isn't about locking it away forever; it's that current cyber defense mechanisms are still too easily 'jailbroken' or bypassed. We need to wait until defense mechanisms are confident enough to handle it before releasing it alongside robust safeguards.
Host: But there's been a lot of pushback: researchers claim cheaper open-source models can replicate it, and others say it's all grand PR and marketing. How do you respond?
Dario:The claim that it can be replicated with open-source models is utterly absurd. The Mythos model discovered 271 new vulnerabilities in the Firefox browser and thousands of new vulnerabilities in many private companies we can't disclose, which previous models simply couldn't find. That person on Twitter who claimed they could replicate it with an open-source model just re-located that specific line of code after we pointed out which line it was. That's not the same thing at all.
As for the 'marketing' claim, not releasing this super-powerful model is a huge commercial loss for us; it could have massively accelerated internal and external R&D. If it helps defenders, it helps attackers equally. So our current strategy is to first give it to defenders to patch system vulnerabilities. Once all the vulnerability 'holes' are plugged, the future internet ecosystem will become extremely secure. Those sniping and doing contrarian takes on X (including certain competitors) fail to take the societal risk trade-offs seriously, showing extreme immaturity.
Host: Have you already made trade-offs you're not entirely comfortable with?
Dario:The entire history of Anthropic is full of trade-offs. In an ideal world, you might spend years studying every pitfall of a chatbot before releasing it. While we did delay the initial version of Claude by a few months, everything is still a trade-off. Now that we are in a leading commercial position, Daniela and I are doing everything we can to nudge the needle further towards 'proceeding cautiously.' That's the intention behind restricting Mythos's release. If you're not a leading player, you can hardly make that decision.
Host: Since this technology is so powerful, why doesn't the government just take you over?
Dario:This is a very serious concern. Throughout history, all unprecedentedly powerful technologies (nuclear weapons, the internet, GPS, cell phones) were initially born in government and federal labs. AI is the first one built entirely in the private sector with the government arriving late, a situation that is actually dangerous and unstable. But the danger of the government not doing it now is arguably greater than the danger of the private sector doing it. So we need checks and balances: we set up the 'Long-Term Benefit Trust,' which has the power to remove the majority of the board and even fire me, introducing an element of public governance.
At the government level, we need the legislative and judicial branches to enforce mandatory pre-release testing and red-line regulation. I fear both a corporate monopoly over it and a government monopoly over it; they must check each other. It's quite ridiculous that there's a group within Silicon Valley who, from being extremely opposed to any regulation or even transparency requirements (saying it kills innovation), upon seeing the first real danger, immediately swing to declaring 'the government should nationalize it.' This reaction of swinging wildly between extreme poles is very immature; what we need is rational middle ground.
Host: Your field always talks about the moment AI gets good enough for 'self-improvement.' Your researchers think it's getting close. How far away specifically?
Dario:I think it's not a specific 'moment' but a continuously accelerating exponential process. We already see AI starting to assist in proposing the architecture for the next generation of AI. A year ago, AI improved total factor productivity by 10-15%; now it might be 20-30%. It won't suddenly go out of control at one moment; instead, on this smooth exponential curve, we need to assess at each node whether we should slow down and apply more control. The sane response is to smoothly escalate our countermeasures as the technology's power grows, not to panic.
Host: You've said there's a 10-25% chance of human civilization collapsing. That's not trivial. Could there be a scenario where something Anthropic builds causes that collapse?
Dario:I certainly hope not. That risk probability stems from the inherent unpredictability of the technology and the complexity of world dynamics. All the measures we take are precisely to reduce, not increase, that collapse probability. It's like the aviation industry: even if you build an airplane ten times safer than others, you can't absolutely guarantee to others it 'will never crash.'
Related reading: OpenAI and Anthropic Collective Change Tune: Is the AI Jobs Apocalypse Narrative No Longer in Vogue?
