Original Title: "Anthropic, Dario Amodei, and the 'Last Private Company'"
Original Author: Insight Beating

Regarding Anthropic's founder Dario Amodei, there are two circulating but contradictory yet equally plausible narratives.

One narrative is that he is the most vigilant watchman of this era. While the world is stepping on the accelerator towards the cliff of superintelligence, only he is willing to stand on the roadside, repeatedly shouting that there is an abyss ahead.

The other narrative is that he is one of Silicon Valley's most ambitious individuals. Leading the world's most powerful AI company, he tirelessly works day and night, gradually crafting the future that everyone has been warned about.

One person can indeed do both of these things simultaneously.

He has given a name to that envisioned world, the "Genius-Land Inside Data Centers." In that country, there are fifty million geniuses who do not sleep, work twenty-four hours a day, think at a speed dozens of times faster than ordinary people, and can replicate countless geniuses like themselves. This country has no land, no borders; it exists within rows of data centers.

In January 2026, he wrote about this imagination in a lengthy article titled "The Adolescence of Technology."

Then he took a step further in his extrapolation. What if one day these fifty million geniuses stopped obeying humans?

The subsequent scenarios became more ominous. Bioweapons, mass unemployment, dictatorship, war – he listed them one by one. Until the end, saying, "We don't have much time left."

Yet in the same article, he also wrote: "AI is already writing a large amount of code for Anthropic, helping to produce the next generation of models more quickly."

Warning on one hand, accelerating on the other. He does not see any contradiction in between.

This is the crux of the whole issue – it's not that he's malicious, he genuinely doesn't see it.

Of course, he has his justification. While Anthropic may pause, OpenAI won't, Google won't, companies from other countries won't either. No one is willing to be the first to let go. The first to halt will likely be unable to stop anything and will merely pass the torch to others.

This rationale can be translated as, since everyone is rushing, let "me" be the one to rush ahead. Since no one can stop, let "me" decide when to pause.

As a result, the question of whether a superintelligence should be created is not so important. In his view, it will eventually be created. What truly matters is who creates it first and, when that day comes, who is qualified to take the lead.

Too Late

Dario Amodei was born in 1983 in San Francisco. His father, an Italian from Tuscany, was in the leather goods business all his life; his mother, of Jewish descent from Chicago, contributed to the planning and construction of the San Francisco Public Library.

From a young age, he had a passion for mathematics and physics. In the late 1990s, during the Bay Area's Internet bubble, companies were popping up one after another, young people were creating websites, seeking investments, and calculating how much they could cash out on IPO day. These stories did not captivate him. Many years later, he said that what he was looking for at the time was "fundamental scientific truths," wanting to understand why the world had evolved into its current state.

He held an almost stubborn seriousness towards what he considered "important issues" from an early age. As long as he believed something was sufficiently important, he couldn't understand how people could just sit back and relax.

In 2003, when the U.S. was gearing up to invade Iraq, he was studying at the California Institute of Technology. There were many anti-war demonstrations on campus. During conversations, some expressed concerns, but the next day, everyone went about their usual routines. He wrote an article in the student newspaper criticizing his classmates, saying that if they were genuinely afraid that a war would cause many deaths, their opposition should not be limited to mere words; if they waited until it actually happened, it would be too late for regrets.

In his view, hesitation was never neutral. Waiting itself was a choice.

What this twenty-year-old could least tolerate was waiting.

Three years later, at the age of twenty-three, his father died of hepatitis C.

At that time, hepatitis C was almost untreatable. The standard therapy involved a year-long regimen of interferon, which caused fever, hair loss, depression, and had a cure rate of less than fifty percent.

In December 2013, a new drug was approved. The treatment duration was shortened to twelve weeks, with a cure rate of ninety-five percent.

His father did not live to see this.

Just seven years short.

A son who couldn't save his father spent the next twenty years urging science to move faster, even faster.

After his father's death, he shifted from theoretical physics to biophysics, researching the neural circuits of the retina at Princeton. At that time, the equipment could only capture a small amount of neural signals, and he, feeling frustrated by the slow progress, joined the effort to develop new sensors.

Later on, he entered Stanford Medical School to study cancer and proteins. The closer to the disease, the smaller a person seems. A disease involves hundreds of genes, tens of thousands of proteins, and countless interconnected pathways. Researchers may spend many years trying to understand one part of it, yet still not know what will happen when they put it back into the entire body.

Later, he described that the complexity almost drove him to despair.

AI then entered his field of view at that moment, as a tool that could expand the scale of human understanding. Machines can read many papers simultaneously, process vast amounts of data, try many possibilities, and replace the cognitive labor that originally required hundreds or thousands of researchers working together.

Starting from the premise of "humans are too small," he arrived at the conclusion of creating something larger and compressing humanity once again.

In 2014, he left life sciences and joined Baidu's artificial intelligence team in Silicon Valley. The team quickly discovered that with a slightly larger model, more data, and additional computing power, the performance would improve further. A few years later, this set of experiences was summarized as the "Law of Large Models." You may not know what the next generation model will suddenly learn, but you can roughly determine that it will be stronger than the previous one.

For the first time, artificial intelligence seemed like it could be mass-produced like an industrial product. Powerful AI thus became something that would inevitably be created sooner or later.

The only differences left were how much data, how many chips, how much power, and how long one had to wait.

The person who hated waiting the most personally proved that the path to the answer he desired did exist.

Steering Wheel

In 2016, Dario Amodei joined OpenAI.

In the following years, language models grew rapidly. By the time GPT-3 was developed, OpenAI allocated over half of the company's computing power to Dario, for him to scale up the model a hundredfold. The results were even more astonishing than many anticipated; it could write articles, translate, summarize, and even had some abilities it hadn't been explicitly taught and learned on its own.

With each increase in scale, there was a corresponding increase in capability.

He was convinced that this path led to general intelligence. And once it was acknowledged that the model might eventually be as smart as a human, safety could no longer be dismissed as trivial.

Then he encountered a problem. Dario was responsible for creating the model, but slowly realized that building the model and deciding where the model was headed were two completely different matters. When the model was released, to what extent it was open, which companies it collaborated with, who had decision-making power—these matters did not automatically fall under his responsibility just because he was in charge of training.

Within OpenAI, there were two factions. One wanted to quickly push the technology out into the world; the other was concerned that who leads the company and the principles by which it operates may be more important than any particular technological detail.

So what kind of person is qualified to lead a company building powerful AI?

His answer is that leaders must be honest, their motives sincere, and they must genuinely desire to make the world a better place.

Technical safety slowly morphed into a judgment of people. This line of reasoning is very dangerous and yet very natural. The danger lies in simplifying all complex issues into one question: Can this person be trusted? And the missing piece in this reasoning is always who gets to make the judgment on whether this person can be trusted.

By 2020, the conflict between him and the OpenAI leadership had become irreconcilable. In December of that year, he left OpenAI with his sister Daniela, Jack Clark, Chris Olah, and several colleagues. Because Dario liked pandas, this small group was internally referred to as the "Panda Faction."

Several months later, they founded Anthropic. Anthropic, by definition, means "pertaining to humans."

This time, Dario was not planning to hand over the reins to someone else.

Twenty People

In 2021, the pandemic in San Francisco had not yet ended.

Having just been established, Anthropic consisted of only fifteen to twenty people. Most of the time, they held meetings over Zoom; and once a week, they would each bring a chair and meet in the park. While having lunch, they discussed models, computing power, security research, and what direction the company should take.

With fewer than twenty people sitting on the grass in San Francisco, perhaps still holding half-eaten salads, their concerns were about the future of eight billion people.

The law of scale told them that the future was inevitable. Since it was bound to happen, someone had to pave the way in advance. Those four words "someone had to" were the entirety of their legitimacy argument.

The goal Anthropic set for itself was to develop cutting-edge models, study how to understand and control them, and then turn this set of security methods into the industry standard for other companies.

Dario had learned his lesson from OpenAI. He understood that only discussing safety on the periphery of technology would likely only result in giving advice to those who truly hold the models. No matter how nicely it is put, it is akin to standing on the shore and teaching someone how to swim in the water.

Getting one's feet wet.

That made Anthropic seem awkward from day one. To understand the dangers of the most powerful model, you had to possess the most powerful model; to influence regulation, you had to have clout in the industry.

So, Anthropic quickly got caught up in the same cycle that all cutting-edge large-model companies couldn't escape: fundraising, buying computing power, training models, releasing products, snatching customers, then using new revenue and valuation to secure more computing power.

Dario said that growing the company was about protecting the world. Investors were also willing to believe that betting on Anthropic meant backing a more responsible AI company.

The cleverness of this argument was that the more money he made, the more ethical he seemed; the higher the company's valuation, the more evidence there was of his "world-saving."

Business interests and a civilizational mission shored each other up.

The direction came from Dario. He wrote long essays on civilization, science, and the future of humanity. Grounding these grand judgments was his sister Daniela, who handled recruitment, team building, and fundraising.

The starting point for the company was an almost unfalsifiable grand judgment:

Powerful AI was on the way, and the world needed a more trustworthy company to be at the forefront.

"More trustworthy," as the company itself put it.

This narrative gave Anthropic a strong cohesion. Those who joined were not just looking for a higher-paying job. They believed that what they were doing was a matter of human civilization.

The most frightening state of a company is when employees truly believe they are saving humanity. Because from that moment on, all business expansion automatically becomes a holy war; all dissenting voices automatically become a betrayal of humanity.

In the first few years of the startup, Anthropic wanted to prove that an AI company could be more cautious and more responsible than others. This desire was probably sincere.

But sincerity never guarantees correctness. Sincere people who do wrong are often more troublesome than hypocrites because they never stop to ask themselves:

What does it really mean to be responsible?

Why should a private company make decisions for the whole world?

Yardstick

Early large-scale model training often relied on a large number of human annotators to judge responses. What kind of answers are better, what kind of content is harmful, whether the model should reject, all had to be judged by humans first, and then these preferences were fed back to the machine.

This method is effective but cumbersome. Annotators come from different places and have different values. No matter how detailed the rules are, they cannot cover all situations.

Anthropic came up with another way. Instead of having humans repeatedly tell the model how to answer each question, it is better to first give it a set of principles. In 2022, Anthropic proposed "Constitutional AI." The approach is roughly to let the model generate answers first, then have it identify the flaws in its own answers based on a set of written principles, find issues, and then make corrections.

This "constitution" is getting longer and longer. It needs to understand power, deal with conflicts of interest, know what maturity and wisdom mean. It even has to think about its identity, situation, and well-being, not simply view itself as a tool with no intrinsic value.

What it is shaping is a kind of personality.

In the constitution, Anthropic writes that this document directly shapes Claude's behavior, representing the company's vision of Claude's values and character, and is the "ultimate authority" on how Claude should act. Historically, those in charge of such matters included the church, emperors, and political parties. Now it is a private company.

But when the impact of a large model on the world is no longer as simple as a one-time consumption. What Claude rejects, what it encourages, how to describe a country, a war, or a political system, will all affect downstream products and become part of how others understand the world.

Anthropic is still making decisions in the manner of a company, but the impact of these decisions is increasingly resembling a set of public institutions, yet without bearing any public scrutiny.

After writing a constitution for the model, Anthropic began designing industry-specific traffic rules. In 2023, the company introduced the "Responsible Deployment Policy." A model that can only write emails and a model that can design biological weapons, launch complex cyber attacks clearly cannot be governed in the same way. So Anthropic delineated different risk levels for the models. The stronger the capabilities, the greater the potential harm they can cause, and the company must implement stricter security measures.

A private company has thus played multiple roles at the same time. It builds models; once the models are built, it measures how dangerous they are; it sets the standard for what is considered dangerous; it establishes the boundaries of when the danger should be stopped; and finally, holding onto this set of results, it then informs regulators on how to set rules.

Being both a player and a referee.

Anthropic clearly had this in mind as well. In order to prevent the company from being entirely swallowed up by commercial interests in the end, Anthropic also designed a long-term interest trust. This trust, composed of external members who do not hold direct economic interests in the company, is intended to ensure that Anthropic, in addition to shareholder returns, also considers the long-term interests of humanity. According to the original design, the trust would gradually acquire the power to appoint and dismiss the majority of directors, and apply the brakes when the company deviates from its mission.

From a corporate governance perspective, this kind of arrangement is quite rare. While most startup companies are busy proving to investors that they will grow rapidly when they raise funds, Anthropic has taken a different approach, setting up a structure to prevent future investors and management from disregarding safety commitments in pursuit of profit.

It cannot be said that these are just empty gestures. Anthropic does indeed take the issue of power more seriously than the vast majority of tech companies, and is willing to cause trouble for itself. However, everything it has thought about from start to finish is how to use this power better, rather than whether or not to have this power.

A company born with a mission will never ask, "By what right do I exist?" It will only ask, "How can I exist better?"

It doesn't answer the question, "Who gave you the power?" Instead, it only declares that since it already has the power, it intends to use it to be a good actor.

At this point, Dario is not just running a company. His philosophy of safety is seeping into the character of the models and into public policy. Next, he begins to discuss which country should be the first to acquire powerful AI, which political systems are worth protecting, the global order of AI acquisition, and who should determine it.

The U.S. Gets the Future First

In 2024, Dario Amodei writes "Machines of Loving Grace."

This title is taken from a poem. In 1967, poet Richard Brautigan wrote "All Watched Over by Machines of Loving Grace" at the California Institute of Technology:

A poem about the reconciliation of machines and all living things. He borrowed the title to write one of the most optimistic technological prophecies in human history.

Potent AI will become a research team composed of countless top scientists, reading papers and designing experiments day and night, condensing decades of scientific progress into just a few years. Cancer, infectious diseases, genetic disorders may be massively overcome. Humanity might truly understand how depression, anxiety, and addiction occur. Lifespans could be extended, and developing countries could leapfrog industrialization.

He made a bold statement that if strong AI development progresses smoothly, the biomedical advancements of the next 50 to 100 years could be achieved in 5 to 10 years.

It's hard not to think of his father.

His son waited for seven years, waiting for a new drug to be approved that could save others but was unable to save him. He hopes that future generations won't have to wait any longer.

In his ideal world, he truly believes that this technology can reduce disease, suffering, and death. He even specifically discussed how to ensure that the benefits of AI don't only accrue to wealthy nations; he hopes that AI can help develop the Global South and spread technological dividends worldwide.

But when he started talking about how to reach this future, things didn't seem so universally beneficial anymore.

He believes that when strong AI emerges, countries must maintain an advantage. This advantage cannot simply be about a company having a better product or scoring higher on a leaderboard. It needs to involve chips, computing power, energy, supply chains, military, and international alliances.

The U.S. and its allies must secure more advanced semiconductors, build larger-scale data centers, while also blocking strategic competitors from obtaining the chips needed to train the most powerful models. When strong AI truly arrives, it will be used to enhance military and economic advantages, establishing a position that other countries cannot easily challenge.

Sharing will come later.

Medicine needs to be distributed, and chips need to be controlled. His promise of "not making future generations wait" is actually more like "not making qualified individuals wait." And he has the final say on who is qualified.

He even used the term "The Eternal 1991" to describe that kind of state. In 1991, the Soviet Union dissolved, the Cold War ended, and the U.S. and its allies gained overwhelming superiority. He hopes that strong AI can recreate a historical moment like that.

Pay attention to the value of this metaphor. A technology that claims to benefit all of humanity, with its ideal blueprint being the most thorough unipolar hegemony in human history.

The world is for everyone; these four words are written beautifully. It's just that "the world" belongs to the U.S., and "for everyone" belongs to Anthropic.

Then DeepSeek emerged. A Chinese company trained competitive models with relatively limited resources. To Dario, this precisely proves that China has outstanding researchers and engineering capabilities. Once they obtain enough advanced chips again, Chinese companies may train the world's most powerful models.

So, we cannot give the chip to it.

The stronger the opponent, the more it proves that the blockade is correct; when the opponent demonstrates their ability, it instead becomes the reason for the blockade. This logic always holds true because it is a self-fulfilling prophecy.

He divided the future into two possibilities. One is a bipolar world where both sides can train powerful models, turning intelligent competition into a new Cold War. The other is a unipolar world where the U.S. and its allies control the majority of advanced computing power. The West leverages its lead in these years to establish technological, economic, and military dominance and then cements this advantage.

He clearly prefers the second future. Concepts such as benefiting humanity, eliminating the need to wait, and achieving harmony between machines and all things are merely preludes in the face of the "Eternal Unipolar."

He has publicly opposed a certain narrative, expressing his dislike for AI company founders portraying themselves as prophets leading humanity to redemption and cautioning against a few companies using their technological advantage to unilaterally shape everyone's future.

He opposes prophets. Yet, what he does is essentially the same as what the prophets do.

He saw all the dangers, except himself

Let's rewind back to January 2026 and "The Adolescence of Technology."

In this article, Dario Amodei essentially preempted all external criticisms of himself and Anthropic.

He acknowledges that leading AI companies are obtaining a level of power historically held by very few private enterprises. He even boldly suggests that AI companies might use their products to influence users' political attitudes, even engaging in "brainwashing." He calls on the public to keep a close eye on AI companies, and governments must also be wary of these companies becoming too entwined with state power.

He also worries that AI will further concentrate wealth and political power. A few companies will hold intelligence, wealth, and information, gradually diminishing the influence of ordinary citizens on the economy and government.

He sees all these dangers. His writing is clear and precise. Few individuals standing at the center of power would candidly admit how dangerous their position is.

It is precisely because his arguments are so compelling that we must ask: What then?

Each time he proposes a solution, it always ends up circling back to the same direction. Following his proposed solutions invariably results in Anthropic gaining more power. The more dangerous the models become, the more it is necessary for Anthropic to stay at the forefront; the less understanding the government has of technology, the more it must rely on Anthropic's judgment; the more easily the public is influenced by the models, the more Anthropic must ensure that the values of the models are well defined; the less trustworthy other countries are, the more the U.S. and its AI companies should hold onto chips and computing power.

He admits that for every ailment, he prescribes the same remedy, Anthropic.

He believes sincerely that this technology is too dangerous to be casually handed over to anyone. He also believes that he saw this danger sooner than most.

A sincere person is sometimes the first to deceive themselves, especially when the benefits of self-deception are so significant.

He seems to sincerely believe that this technology is too dangerous to be casually handed over to anyone. He also sincerely believes that he saw this danger sooner than most.

Putting these two sentences together forms a complete self-authorization. He never has to declare himself the guardian of humanity. He just needs to continually eliminate those he deems unreliable: the market is inadequate, the public is inadequate, authoritarian states are inadequate, other companies are not cautious enough, and when all these are completely eliminated, the one best suited to steer the ship happens to be Anthropic.

The ultimate end of this process of elimination is always himself.

He does not believe anyone is qualified to decide the future of humanity on behalf of others.

So, he has decided to do it himself.

The Last Private Company

On August 14, 2026, a statement was uttered on a podcast.

"Within Anthropic, they are very confident. Multiple people I trust have told me that Dario once said: Anthropic may one day become the only private company in the world."

The speaker was Gavin Baker, a hedge fund manager who invested in SpaceX. He added that in that vision, there would only be Anthropic, followed by various national governments. Only these.

Within a day, this statement had spread across social media.

Former White House AI and cryptocurrency advisor, currently Co-Chair of the President's Technology Advisory Board, David Sacks, remarked, "This is too arrogant. It's a bit like SBF's audacity."

The following evening, Dario published two lengthy articles on X in response.

The first article discussed regulation. He stated, "The 'either centralize or distribute' dichotomy is a false choice; a truly good system can actually decentralize power:"

「The formal court system may sometimes appear outdated and elitist, but it is better at protecting the vulnerable than the alternative—mob justice. At its best, the system can empower ideals over individuals.」

The second piece discusses the transmission of information. He said that public's negative sentiment towards AI is a trust crisis. He then wrote the most impactful section of the two lengthy articles:

「I believe that so far, the most accurate criticism of AI companies including Anthropic is that we have not yet delivered on our grand promise to benefit the world. This is entirely our fault.」

「The truly effective approach is to genuinely cure cancer.」

Both lengthy articles, not a single sentence addresses that rumor positively. He used a real question to dodge another real question.

He said, to truly cure cancer. By this, he means using AI to drive overall medical advancement, to get new drugs out fast enough, faster than any one disease.

This is a quote for someone who died in 2006. That person waited for a drug, waited long, and never got it.

Since then, he believes slowness is a sin. If one believes slowness is a sin, they will take speed as the only remedy. Speed is a good thing. It's just that he wants everyone to believe that the only cure can be dispensed by him.

In 1986, British writer Alan Moore published a comic called "Watchmen," which was later made into a movie. In the movie, there is a character named Adrian Veidt, also known as Ozymandias, claiming to be the world's smartest person.

He saw the impending disaster earlier than everyone else. Nuclear war, doomsday, human self-destruction. He spent half his life preparing for it. Then he realized that merely warning was useless.

So he personally triggered that disaster. Overnight, several cities were wiped off the map.

He felt that only by letting humanity see the face of destruction with their own eyes would the two superpowers stop, and the world could finally have peace.

He was the perpetrator, but also the savior he acknowledged himself to be.

At the end of the story, everything goes as he wished. He asks that blue, near-omniscient Doctor Manhattan:

「The end? Adrian, nothing ever ends.」

Original Article Link

Welcome to join the official BlockBeats community:

Telegram Subscription Group: https://t.me/theblockbeats

Telegram Discussion Group: https://t.me/BlockBeats_App

Official Twitter Account: https://twitter.com/BlockBeatsAsia