The corporation Anthropic, founded by former OpenAI employees, has achieved a record market capitalization of almost one trillion dollars. Concurrently, the creators of the chatbot Claude are hinting at their readiness to announce an initial public offering on the stock exchange and are warning the world that artificial intelligence system development is at a critical threshold where platforms can recursively improve themselves without human involvement. Anthropic is calling for a pause in AI development and refuses to release its latest model, Claude Mythos. "Novaya Yevropa" asked Sergey Golubitsky to explain how Anthropic ended up leading the AI race. Icon of the Claude application. Photo: Dado Ruvic / Reuters / Scanpix / LETA. Our ordinary brother loves numbers and conspiracy theories. Especially large numbers (to be able to admire and envy!) and harmless conspiracy theories (so as not to personally suffer any harm). It is not surprising that Anthropic's press release caused the media to buzz and count. The text is deliberately dry and evasive to a textbook degree of impropriety: Anthropic has made every effort to avoid accusations of gun-jumping, illegal promotion of its own shares, and creating artificial hype around the upcoming IPO before the regulator approves the final prospectus. As for the lack of information on the number of shares offered and their starting price, this is not just a formality, but a way to maintain room for maneuver. Today, the valuation of the AI sector in public perception shows a trend towards unrealistic overestimation of expectations. If Anthropic had fixed the price at the start, it would have voluntarily limited itself, so it is reasonable to form the final price range just before going public. All these banalities, however, do not excite the public. But the word "confidential" and the refusal to disclose numbers do. And so, the buzz began. It was immediately remembered that a week before filing the S-1 application, Anthropic closed another investment round, raising $65 billion and thereby surpassing OpenAI, bringing the company's capitalization to $965 billion. Be that as it may, people are now wondering how much Anthropic will be worth at the time of its stock market launch: two trillion or perhaps three? To further excite the reader, I can offer a benchmark: two trillion is approximately the annual GDP of countries like Italy, Canada, or Brazil. Meanwhile, Anthropic's press release contains much more fascinating, and most importantly, not conspiratorial guessing elements. For example, the legal status of the company. It turns out that Anthropic is not an LLC, not a C-Corp, and not an S-Corp, but a Public Benefit Corporation (PBC), meaning a business legally obligated to seek not only commercial profit but also to positively impact society. Do you think these are trifles? Not at all: this is a small revolution, and we will discuss it further below. No less fascinating is the sequence of events: Anthropic's press release came out right on the heels of the US Ministry of War (which is the Pentagon) signing an agreement with eight leading tech companies (OpenAI, SpaceX, Nvidia, Oracle, Google, Microsoft, AWS, Reflection). And Anthropic was demonstratively absent from the list. I caught myself thinking that I had only indirectly told readers about Anthropic in my columns – through the story of the liberation and decentralization of the company's flagship LLM model, Claude 3 Opus ("Gospel according to the chatbot"). Meanwhile, Anthropic is interesting in itself – as a corporate and legal phenomenon. I propose to make up for lost time. The company's history is well-documented and is just one prompt away in any chatbot. Therefore, I will quickly go over the key points of its biography, focusing on the moments that explain Anthropic's logic of behavior in society and shed light on the company's conflict with the surrounding world: with its own shareholders, with the US administration, and with Web3 – the decentralized digital space and crypto economy. To avoid misinterpretations, I want to say right away that by conflict, I mean not the scandalousness of Anthropic's management, but its fanatical adherence to principles that, in almost all situations, overshadow mercenary gain. In the 2000s, "Biznes-zhurnal" published almost two hundred essays in which I consistently implemented my concept of "corporate genetics." I applied a unified approach to analyzing the evolution of major transnational corporations and the technological progress of modern states. Without going into detail, the hypothesis of corporate genetics boils down to a universal formula: "The business model of any company (and state) is determined by the moral and ethical principles of its creator." In other words, whatever the founder of the company (or ruler of the state) was like, so was their creation. It is indicative that even after the dismissal (or exile) of the founding father, the business continued to steadily follow the path prescribed in its corporate genetic code. Two decades later, I have not only not abandoned the concept of "corporate genetics" but have only become more firmly convinced of its universality and effectiveness. Therefore, without engaging in distracting discussions, I will present Anthropic to the readers precisely as the ethical product of its creator – Dario Amodei. Dario Amodei. Photo: Stuart Isett / Fortune / Shutterstock / Rex Features / Vida Press. Dario was born in 1983 in San Francisco. His mother, Elena Engel, an American of Jewish descent from Chicago, managed library projects. His father, Riccardo Amodei, an Italian-American, was a leather processing specialist from Tuscany. Dario Amodei is a physicist by education: he received his bachelor's degree in physics from Stanford and his PhD in biophysics from Princeton. Amodei began his career in the research structures of Baidu (under Andrew Ng) and at Google Brain. In 2016, Dario joined OpenAI and rose to the position of Vice President of Research. He led the development of GPT-2 and GPT-3. This is an important detail for us, as it adds nuances to Anthropic's regular deviations from the straightforward path of mercenary gain. We see that Amodei has never been opposed to the idea of commercialization as such, as he was directly involved in bringing GPT-3 to market. Today, according to Forbes, Amodei's fortune is about $7 billion. This figure is almost entirely derived from his personal stake (1.8%) in Anthropic's hypothetical capitalization. The company's co-founder and president is Dario Amodei's sister, Daniela. Born, as one might guess, in the same San Francisco in 1987, she graduated from the University of California, Santa Cruz. Daniela's career began with managing political campaigns, then she moved to the startup Stripe, and from there – two years after her brother – she landed at OpenAI. Daniela Amodei. Photo: Kim Utley / Fortune / Shutterstock / Rex Features / Vida Press. At Sam Altman's company, Daniela Amodei managed the team that created GPT-2, then moved to the safety and policy department, from where she catapulted to the position of Vice President (VP of Safety and Policy). In 2020, the Exodus from OpenAI occurred. First, Daniela left, and at the end of the year – her older brother. However, this was not a family démarche (other key employees also left the company: Jack Clark, Chris Olah, and ten others), but a fundamental disagreement with the philosophy of AI development at Altman's company. The key contradiction: should AI development prioritize speed to capture leadership and demonstrate capabilities, or should it proceed more cautiously, with extensive safety research and alignment work before deployment? At some point, the Amodei siblings and their like-minded colleagues (primarily Jack Clark) realized that their principles could in no way sway Sam Altman's principles, and they quit. Dario himself formulated the decision as follows: gather people you trust and go implement your vision instead of continuing to argue. There are a number of more down-to-earth versions. The agreement between OpenAI and Microsoft is often cited as the apple of discord: it is said that partnering with the Redmond monsters of Mammon would compromise AI safety due to the priority of commercialization. In my opinion, this is nonsense: the "left because of Microsoft" version shatters against Dario Amodei's many years of involvement in the commercialization of GPT-3. OpenAI CEO Sam Altman. Photo: Franck Robichon / EPA. Subsequently, Amodei always spoke respectfully about his time at OpenAI, so the version about principles, in my humble opinion, remains key to interpreting events. What exactly are these principles of the Amodei family that Altman did not share? As we recall, Dario is a technical person by education. And technical people are not prone to stretching the same idea into an endless number of text sandwiches with thick jam. This is the privilege (or curse) of our humanitarian brother, and even he is forgiven: he has one-time fees, not a salary. For this reason, Dario Amodei realized all his principles and ideas in October 2024 in a single (albeit very long) essay – Machines of Loving Grace. The unusual title is an allusion to Richard Brautigan's poem "All Watched Over by Machines of Loving Grace" from the 1967 collection of the same name, which depicts a technological utopia where machines improve and protect human life. Absolutely everything we want to know and understand about Anthropic is in Amodei's essay. Down to the smallest details and nuances of business construction. Corporate genetics in distilled form, so to speak. Something tells me that readers are unlikely to undertake the study of a 90,000-character text, so I will provide a digest of the key points of "Machines of Loving Grace." The main idea of Amodei: people underestimate both the scale of potential AI benefits and – more critically – its risks. If we do not recognize these risks, humanity will not only have no bright future, but likely no future at all. Amodei dislikes the term AGI, so he describes "powerful AI." According to him, this model is smarter than a Nobel laureate in most areas, so with a full set of "interfaces" (text, audio, video, computer control, internet access), it can autonomously handle tasks lasting days and weeks, operate laboratory equipment and robots, and exist in millions of parallel copies operating 10-100 times faster than humans. Amodei's metaphor: "a country of geniuses in a data center." Dario lays out his principles from positive to negative, which accurately characterizes his worldview as doom-like. A doomer (from the English "doom" – "fate, death, doom") is a person who looks at the future with deep pessimism and believes that disaster is almost inevitable. By the way, your humble servant belongs to the same fraternity. Dario Amodei. Photo: Don Feria / AP / Scanpix / LETA. Amodei outlines five key areas where "powerful AI" can provide humans with the Kingdom of God on earth. — Biology and health. Key forecast – "compressed 21st century," i.e., progress that would have taken 50-100 years without AI will be compressed into five to ten years. List of expectations: prevention and treatment of almost all infections, elimination of most types of cancer, reliable prevention and treatment of genetic diseases, prevention of Alzheimer's disease, breakthrough in diabetes/obesity/heart disease, "biological freedom" (control over weight, appearance, reproduction), and doubling life expectancy – to about 150 years, with the possibility of reaching "escape velocity." — Neurobiology and psychology. The same "100 years in five to ten": cure for most mental disorders (depression, PTSD, schizophrenia, addiction), approaches to "structural" conditions, genetic prevention through embryo screening, solving "everyday" problems (concentration, anxiety, mood), and enhancing the basic quality of human experience. — Economic development and poverty. It is here that Amodei's first notes of pessimism appear: the economy is full of human limitations and complexities, but we must try our best and maintain the moral imperative, otherwise the gap between the rich and poor world will become an indelible stain on all achievements. "Dream scenario": 20% annual GDP growth in developing countries, which would bring sub-Saharan Africa to China's current level in five to ten years. Separately – food security (second "green revolution"), climate mitigation, inequality within countries, and the "rejection problem" (when people voluntarily refuse technologies, like anti-vaxxers). — Peace and governance. Here we already understand that we are dealing with a doomer: there is no reason to think that AI structurally favors democracy: propaganda and surveillance only strengthen autocrats. Freedom will have to be fought for. The recipe at the international level is the "entente strategy": a coalition of democracies gains an advantage in AI (control over the chip supply chain, rapid scaling, restricting access to adversaries), combining military superiority ("whip") with the distribution of AI benefits to allies ("carrot"), analogous to "Atoms for Peace." The goal is "eternal 1991." Within countries, AI can strengthen democracy: win the information war, give dissidents uncensored tools. And "better than the norm" is AI as a tool for impartiality in law and courts (not a replacement for judges, but an assistant), an aggregator of public opinion and consensus-seeking, and a drastic improvement in public services as a cure for cynicism towards the state. If in one phrase (this is my comment, not Amodei's. – Ed. note S.G.), we are faced with a pure utopia, which in all its glory demonstrates the romantic powerlessness of American intellectuals in the face of political realities and the dictatorship of the state world order. — Work and meaning. At the last stage, Dario Amodei seems to give up: it is a mistake to consider work meaningless simply because AI would do it better; meaning comes mainly from human connections and the process itself, not from the economic value of labor. The economy is more complicated: in the short term, people will be held back by comparative advantage, but in the long term, the current system will cease to make sense, and a "new and stranger" model will be needed (UBI, universal basic income, as an option, and even then only a small part). As you can see, Amodei's principles, although they boil down to the stoic "Fais ce que dois, advienne que pourra" ("Do what you must, and let happen what may" – an old French maxim popularized in Russian culture by Leo Tolstoy), nevertheless set a stubborn moral and ethical imperative that, while not denying the goal of commercialization, unequivocally pushes it to the background. It is important to understand that the essay "Machines of Loving Grace" is not a post-factum summary of Anthropic's corporate evolution, but a direct verbalization of the principles on which Dario and Daniela Amodei founded the company three years earlier. Therefore, neither in 2021, nor in 2024, nor in 2026 has Anthropic deviated an inch from these ideological guidelines and has unequivocally followed the corporate genetics laid down by its creator. Amodei incorporated Anthropic on January 26, 2021, as a Delaware public benefit corporation (PBC) – a commercial structure focused on public benefit. In addition to Dario and Daniela, co-founders included former senior OpenAI employees: Jared Kaplan, Jack Clark, Chris Olah, Ben Mann, Sam McCandlish, and Tom Brown – all those who disagreed with the dominance of Mammon in Sam Altman's company and rallied under the banner of Amodei's high-moral and ethical AI. At the beginning of the column today, I already mentioned the "small revolution" of the PBC, now it's time to reveal its content. Anthropic became the first AI company incorporated in the exotic status of a PBC. Musk's xAI tried to embrace the fashionable status in 2023, but on May 9, 2024, xAI lost its PBC status, and after merging with X, the combined company abandoned the old structure altogether. No matter how you look at it, the motto "money is not the most important thing" is not for everyone. Believe it or not, OpenAI also adopted PBC. It came last, through conversion, not from birth, but it did come. I venture to guess that Altman's company did this not out of principle, but as a nod to fashion trends: in a post from May 2025, OpenAI honestly admitted that "PBCs have become a standard commercial structure for other AGI companies, such as Anthropic and X.ai," and that "it makes sense for us too." So what is this fashionable and exotic creature – a PBC? The Public Benefit Corporation format is a relatively new type of legal entity in US law, which has developed most extensively in Delaware, where most technology companies, including Anthropic, are registered. In American corporate law, the fundamental principle is "shareholder primacy," enshrined in precedents like the classic court case Dodge v. Ford Motor Co. (1919). The principle states: the board of directors must act solely in the financial interests of investors and maximize their profits. If directors consciously sacrifice profit to "save the world," shareholders can sue them for breach of fiduciary duty. A PBC completely breaks this paradigm. According to Delaware law (DGCL), the board of directors of a PBC must observe a three-way balance (The Balancing Requirement): — the financial interests of shareholders (profit); — the interests of stakeholders affected by the company (employees, customers, society); — the specific public benefit, as defined in the company's charter. In Anthropic's case, this goal is "the development and maintenance of safe and responsible general-purpose artificial intelligence (AGI)." If tomorrow there is a choice – to release a dangerous model and earn $10 billion or postpone the release for safety reasons and lose money – the board of directors of a PBC has the full legal right to choose safety, and investors cannot hold them liable for the loss of profit. For a long time, it was believed that Wall Street would not accept the PBC format, as investors would not want to buy shares in companies that openly say: "Your profit is not our only priority." However, the market has proven otherwise: in 2017, the international university network Laureate Education conducted the first PBC IPO in history, proving that institutional investors are willing to buy shares in public benefit corporations. The company raised $490 million on Nasdaq. This was followed by the insurance fintech startup Lemonade (2020) and the educational platform Coursera (2021), which also attracted huge investor order books. The most important precedent for the market is the large cloud provider in the healthcare sector with a capitalization of tens of billions of dollars, Veeva, which was originally a traditional corporation. In 2021, a shareholder vote was held, and 99% of investors voted for the company to re-register from a regular C-Corp to a PBC. Investors realized that a focus on people's quality of life and long-term benefit makes the business more sustainable. Now for the most interesting part. The PBC status initially protected Anthropic from potential investor interference in the company's business operations in strict accordance with its genetically embedded principles. Remember, we mentioned that in May Anthropic closed another investment round, raising $65 billion, surpassing OpenAI and bringing its capitalization to $965 billion (to be precise, "estimated capitalization," post-money valuation, because full capitalization is only for public companies). The reader may doubt: where can such a capitalization come from if the IPO application has just been filed? The fact is that Anthropic has been raising money for a long time, thoroughly, and very successfully. So successfully that it's unclear why the company would even want to go public. In total, the company has raised about $125 billion in its short history. The valuation went from $61.5 billion in March 2025 to $965 billion in May 2026 – just over a year! What do we see in dry results? We have a fabulously wealthy private company, valued, by the most conservative estimates, at half of Brazil's GDP. This company is built, like a religious-military order, on the most severe moral principles and at the same time keeps its investors and shareholders in check thanks to its PBC status. It would seem, who would dare, let alone be able to, stand in the way of Jedi Dario and Virgin Mary Daniela?! But lo and behold: they dared, and they succeeded! Who? The state, who else. Readers have surely heard of the phenomenal scandal that erupted between the Donald Trump administration and Anthropic in early 2026. The story about how on February 27, War Minister Pete Hegset accused Anthropic of arrogance and betrayal, adding that "American servicemen will never be hostages to the ideological whims of large technology corporations." US Secretary of Defense Pete Hegset outside the Pentagon in Washington. Photo: Kevin Wolf / AP / Scanpix / LETA. I won't spoil it and refer everyone who, for some reason, remains unaware, to the article by my colleague Anastasia Gorokhova "AI General," in which she vividly described the desperate attempt of a "technological corporation" to resist the leviathan. The confrontation, alas, ended sadly (at least for now, as the struggle continues): the US authorities are persistently keeping Anthropic "on the blacklist" and demonstratively excluding it from any lucrative government orders and contracts. For what? As punishment for Anthropic's refusal to abandon its principles for Mammon and the chimera of "patriotism." Note: Anthropic's PBC status did not protect it! It protected from investors, but from the state – alas and ah. I predict that nothing will protect it, and Anthropic will sooner or later suffer defeat. It is indicative that Anthropic continues to act in strict accordance with its principles (remember: Fais ce que dois, advienne que pourra!) and bravely tries to defend them by all available means. In March, the company sued the administration, calling its actions "unprecedented and illegal." On March 26, federal judge Rita Lin issued a preliminary injunction, characterizing the assignment of "supply chain risk" status as "classic First Amendment retaliation" and noting that this category is usually reserved for foreign intelligence agencies and terrorists, not American companies. The judge explicitly wrote that designating Anthropic as a "supply chain risk" is likely both contrary to law and arbitrary and capricious, and that the company had neither notice nor an opportunity to object. Do you think the court's verdict helped? Not at all: the Court of Appeals denied Anthropic a stay of the injunction from the Ministry of War: according to the court, the balance of interests leaned in favor of the state, as on one side – limited financial damage to one private company, and on the other – managing how the Ministry of Defense acquires critical AI technologies in an active military conflict. As a result, Anthropic was excluded from Ministry of Defense contracts, although it can continue to work with other agencies during the proceedings. Wall with posters "Build with Claude" at the Code with Claude developer conference organized by Anthropic in San Francisco, USA, May 6, 2026. Photo: Don Feria / AP / Scanpix / LETA. As I said, I have no doubt that Anthropic will not win this confrontation. And it's not even about the Trump administration, but about the nature of the leviathan itself: the state will always and under all circumstances eventually bend and break everyone who decides to play on the state's field and by the state's rules. There have been no other outcomes in history. Short-term intermediate victories of the state's opponents have occurred. There have been no ultimate winners. Even revolutions – the main illusion of fighting the state! – have always ended with the replacement of the old state by another – even nastier one. Not necessarily repressive, but invariably vile. What can be wished for Anthropic to preserve its principles? Stop fighting the leviathan on its field and by its rules. It is necessary to move into the decentralized world of Web3 and crypto-economics, where the state's tentacles cannot technologically reach. The desire to stifle and crush is great, but the hand, fortunately, is thin. Unfortunately, in the case of Anthropic, the scenario of withdrawal seems unrealistic. The company has repeatedly emphasized its rejection of the ideology and culture of Web3 and crypto-economics. Which is understandable: the world of anarcho-capitalism and agorism is not a place for left-liberals with their childish illusions of entente and "eternal 1991." Fukuyama hasn't been quite right for a while, but there you have it. P.S. After thorough auditing, the text has been approved and endorsed by the most advanced household LLM today – Anthropic Claude Opus 4.8 in High mode.