A pivotal split that will reshape the AI landscape! The hidden grievances between OpenAI and Anthropic

In December 2020, the last few minutes of an all-hands video conference at OpenAI completely changed the landscape of the AI industry.

Jack Clark, then OpenAI's policy lead, typed "I love you all" in the chat box, and the screen immediately went black—seven core OpenAI members were simultaneously kicked out of the meeting, their email access revoked, their Slack accounts deleted, and their laptops remotely locked. This was not an accident, but the end of a long-planned split.

These seven people were taken away by Dario Amodei. At the time, he was the Vice President of Research at OpenAI, a key driver of GPT-2 and GPT-3, and the one who wrote a 26-page paper internally, laying the foundation for the large-scale model technology roadmap. His reason for leaving was not money or position, but because he believed that OpenAI's management—especially Sam Altman—could not be trusted at all.

On September 29, according to Kevin Roose, a reporter for The Atlantic, after years of research and interviews with more than 150 people, the root of this rift can be traced back to an internal plan known as the "Countries Plan" in 2017, and the suspicion, quarrels and feelings of betrayal that accumulated over the next three years.

Five years later, Anthropic, founded by those seven people, has thousands of employees and is racing alongside OpenAI toward AGI. But according to multiple sources, the relationship between Dario and Altman remains one of mutual disdain .

That 26-page paper, and Amodei's ambition

In 2016, Dario Amodei left Google to join OpenAI.

He wears round-framed glasses and has a head of brown curly hair. When he's nervous or thinking, he has a habit of twirling the curls in his hair with his fingers. His colleagues say he's "a genius, but stubborn."

Dario Amodei video screenshot

His way of expressing his views is by writing long articles.

In 2017, he wrote a 26-page internal paper titled "The Big Blob of Compute." The paper proposed that machine intelligence is essentially a function of scale—the larger the model, the smarter it is.

This judgment directly propelled OpenAI's research towards Large Language Models (LLM), and laid the foundation for GPT-2 and GPT-3.

Logically, Amodei's career should be at its peak. His team not only proposed "Scaling Laws" to improve the performance of predictive models, but also pioneered "Reinforcement Learning Based on Human Feedback (RLHF)" technology to keep models away from harmful outputs.

But he became increasingly convinced that he would not stay to develop GPT-4.

"Countries Plan": The plan that completely disheartened him.

Amodei has lost faith in the company's leadership.

He considers Sam Altman a "two-faced manipulator" and President Greg Brockman "power-hungry and reckless." He disapproves of their decision to commercialize OpenAI, believing it deviates from the original intention of being "unrestricted by the need to generate financial returns."

What disgusted him even more was the internal factional struggle.

A person who had spoken with Amodei revealed that Amodei at the time considered OpenAI to be "a farce" and "run by people who lack integrity." This person bluntly stated, "There are new crises every day, and if Dario doesn't throw a huge tantrum, Sam will do something crazy within 24 hours."

The fuse was lit back in 2017.

At the time, Musk was withdrawing from OpenAI, and the company was in dire need of funding. According to five people familiar with the matter, Brockman and Ilya Sutskever devised a fundraising scheme known as the "Countries Plan" : auctioning off the rights to OpenAI's future AGI technology among governments including the United States and other major powers.

Brockman and Sutskever's logic is: AGI has extremely high national security significance and will inevitably be taken away from private companies by the government sooner or later. Since it will be handed over eventually, why not sell it now to raise development funds?

"It sounds like tech people are pitching a crazy startup with a bizarre plan, and the final step is: we'll get the government to foot the bill," recalled one person who heard the plan presented.

Amodei was stunned. He believed that selling AGI to governments ran counter to OpenAI's mission.

In December 2017, he wrote an email to Brockman, Sutskever, and other executives, using direct language:

“In my view, this plan will not only fail to reduce, but will exacerbate, the arms race, potentially leading to human rights abuses… and undermining the moral legitimacy of OpenAI.” He insisted that OpenAI should be optimized for the benefit of humanity.

An OpenAI spokesperson later told The Atlantic that the company "had briefly considered turning OpenAI into an international collaboration," but "never considered selling AGI to any country."

This plan was ultimately shelved due to Microsoft's involvement. But Dario did not forget that Brockman and Sutskever had seriously discussed an idea he considered "morally disgusting."

Trust was gradually eroded: the hidden list, the rejected acquisition terms, and the Kanye West affair.

Over the next two years, the rift widened. Amodei began documenting the moments he believed he had been misled by Altman.

For example, during the private beta testing of GPT-3, Amodei requested a list of approved clients from Altman. Altman sent him one. However, Amodei later claimed that he had found another, longer list of approved clients, which Altman made no mention of.

For example, during negotiations with Microsoft for a $1 billion investment, Amodei strongly insisted on retaining the "merger and cooperation clause" in OpenAI's bylaws—which stipulates that if a safer AGI project emerges, OpenAI should cease competing and assist the other. He maintained that Microsoft had no right to veto a potential future acquisition of OpenAI (which is the most likely way to achieve merger and cooperation).

Altman told him that Microsoft did not agree to the terms. But Amodei was convinced that Altman had conceded (or hadn't mentioned it at all) in order to rush the deal and then lied to him.

Some seemingly absurd business decisions were even more intolerable to Amodei. In 2020, Altman met with rapper Kanye West and proposed that he participate in the GPT-3 beta testing. At the time, Kanye had just released a video in which he appeared to urinate in front of the Grammy Award trophy, and his mental state was extremely unstable. Policy Director Jack Clark ultimately stepped in to nip this dangerous idea in the bud.

These things, taken individually, might not be enough to make someone resign. But in Amodei's mind, they pieced together a conclusion: such powerful technology as AGI must never be handed over to such a person.

The "Plump Toy Advisory Group," and an increasingly isolated safety-conscious clique.

Amodei's sister, Daniela Amodei, was also fed up with OpenAI.

As a manager with a background in politics and high emotional intelligence, Daniela complemented her brother's tech fervor. However, as she took over more technology teams, her conflicts with management increased.

Screenshot from Daniela Amodei video

Interestingly, Daniela has a unique conflict resolution mechanism – a plush toy advisory group.

She and board member Holden Karnofsky collected a large number of plush toys and assigned them roles: Jinji is a lazy cat who encourages her to cancel meetings and do less work; Maura the pink bear is a detail-oriented Type A manager; and Beary Bonds the panda is a "generous and empathetic guardian bear." She uses this semi-joking approach to management role-playing.

By 2020, the Amodei siblings had begun to build their own core circle within OpenAI.

They set up a private Slack channel, filled with security jargon that non-technical people couldn't understand. A former colleague recalled: "Dario had a habit of just saying: We have to do X; it's the morally right thing to do. It was like a shortcut to winning any argument—'Because of security.' This really annoyed the non-security teams."

In 2019, Amodei considered giving up and even considered switching to DeepMind. But Jared Kaplan, a physics professor in charge of scaling research, proposed a more disruptive idea: since their philosophies clashed, why not bring all the security-conscious people and start their own company?

In 2020, the same year GPT-3 was released, Amodei assembled seven co-founders.

Before a complete showdown, Amodei presented Altman with a series of conditions for his continued employment, including granting the security team a "veto power" over the release of new products. He knew this would most likely be rejected, but he hoped to force Altman to compromise.

During the negotiations, the seven individuals created a secret Slack channel called “BATNA”—a business school term meaning “best alternative to the negotiated agreement”—ready to walk away at any time.

When the network cable was unplugged, the screen went black.

After confirming that Altman would not meet the demands, the "BATNA" team hired lawyers and began negotiating terms known as "divorce".

Both parties reached an agreement: the departing team could poach the employees they wanted within a few weeks in exchange for a promise not to poach any employees for the following year; they could take some internal research documents with them; and both parties signed a mutual non-disparagement agreement, promising never to speak ill of each other in public.

In December 2020, a crucial online meeting for all OpenAI employees was held, which would determine the company's fate.

Altman began as usual, reporting on the research progress. He then handed the floor to Amodei.

Amodei calmly announced that he and six other members of the leadership team would be resigning to form a security-focused AI organization. The meeting chat window was instantly flooded with shocked emoticons and exclamation marks. Everyone was speculating: Who will train GPT-4? Who's my new manager?

Amidst the stunned silence, Jack Clark typed his final message in the chat: "I love you all."

Immediately afterwards, in an instant, all seven people's screens went black.

OpenAI kicked them out of the meeting, revoked their email and Slack permissions, and remotely locked their laptops.

A few hours later, this group of newly unemployed founders held their first video conference. No computers, no GPUs, no data, not even a company domain name (originally intended to be Sparrow Systems, but ultimately changed to Anthropic). But they raised their glasses in front of the screen, laughing more heartily than they had in the past few months. They were free.

A trillion-dollar valuation and an unquenchable resentment

Five years have passed, and reality has played a cruel joke on our original intentions.

In his founding vision written in 2021, Amodei envisioned a small AI security research lab with about 50 people and predicted that building AGI would only require about $10 billion.

In the summer of 2023, this place resembled a thriving cult headquarters or a university plant lab. 160 security-obsessed employees huddled together eating vegan sandwiches and stacked high thrones with empty Liquid Death cans.

But by the winter of 2025, Anthropic had become a command center.

The staff swelled to several thousand and moved into a heavily guarded high-rise building. A dozen burly bodyguards stood at the entrance, and even the floor where the new Claude was being trained required special access for its own staff.

Amodei has become a tech giant traveling by chartered planes and surrounded by bodyguards. He was once low-key and reserved, but now he manages one of the fastest-growing capital behemoths in history, racing towards a trillion-dollar IPO.

Daniela Amodei calls it all a “successful disaster”: “There was pressure on every dimension. It was frightening because we could lose it at any moment. That’s just the business side. Then there was our mission, and we were thinking, what if we accidentally made things worse? Like, we thought we knew everything about security, but did we really?”

The only thing that hasn't changed is their bitter rivalry with OpenAI.

According to sources familiar with them, Dario and Daniela still hold Altman and Brockman in contempt. In Anthropic's internal Slack group, Dario even bluntly stated that Altman and OpenAI were "evil," to the point that during the 2023 OpenAI board coup, PR executives even asked him to delete the message for fear of it being leaked.

Meanwhile, at OpenAI, employees privately described the Amodei siblings as "power-hungry signalers of virtue," accusing them of hypocrisy—ostensibly warning of the risks of AI while secretly accelerating the race.

This has become the most profound dark humor in Silicon Valley today: both groups are pouring their computing power into training AI models to maintain kindness, empathy, and restraint; but humans, as the creators, can never let go of their deep-seated hatred for each other.

Risk warning and disclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.