In December 2020, during the final moments of an all-hands video meeting at OpenAI, the landscape of the artificial intelligence industry was completely transformed.

A policy chief typed "I love you all" into the chat box, and then the screen went black — seven core OpenAI members were simultaneously ejected from the meeting, their email access revoked, their Slack accounts deactivated, and their laptops remotely locked.

This was not an accident, but the endpoint of a long-planned rupture.

Those seven individuals were taken away by Dario Amodei. At the time, he was OpenAI's vice president of research, a central driver of GPT-2 and GPT-3, and the person who had written a 26-page internal paper laying out the technical roadmap for large models.

His reason for leaving was not money, not position, but his belief that OpenAI's management — especially Sam Altman — simply could not be trusted.

According to disclosures by journalist Kevin Roose after years of investigation and interviews with more than 150 people connected to the matter, the roots of this rupture can be traced back to a 2017 internal proposal known as the "Countries Plan," and the accumulating suspicion, arguments, and sense of betrayal over the following three years.

Five years later, the Anthropic founded by those seven people now has thousands of employees and is racing neck-and-neck with OpenAI toward AGI. But according to multiple people familiar with the situation, the relationship between Dario and Altman remains one of "mutual contempt."

That 26-page paper, and Amodei's ambition

In 2016, Dario Amodei left Google and joined OpenAI.

He wore round-framed glasses, had a head of brown curly hair, and when nervous or deep in thought, had a habit of twirling a curl around his finger. Colleagues described him as "a genius, but stubborn."

His way of expressing views was to write long documents.

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

This judgment directly pushed OpenAI's research direction toward large language models (LLMs), and also laid the foundation for GPT-2 and GPT-3 later on.

By any measure, Amodei's career was at its peak. His team not only proposed the "Scaling Laws" that predict performance improvements in models, but also pioneered "Reinforcement Learning from Human Feedback" (RLHF), a technique that keeps models away from harmful outputs.

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

The "Countries Plan": the proposal that thoroughly disillusioned him

Amodei lost confidence in the company's leadership.

He regarded Sam Altman as a "two-faced manipulator," and president Greg Brockman as "power-hungry and reckless." He disliked their decision to commercialize OpenAI, believing it deviated from the original intention of being "unconstrained by the need to generate financial returns."

What disgusted him even more was the internal factional strife.

A person who once spoke with Amodei said that Amodei at the time regarded OpenAI as "a farce" run by "people lacking integrity." This person said bluntly: "Every day there was a new crisis. If Dario didn't throw an absolute fit, Sam would do something new and crazy within 24 hours."

The fuse had already been laid in 2017.

At the time, Musk was withdrawing from OpenAI, and the company urgently needed funding. According to five people familiar with the matter, Brockman and Ilya Sutskever conceived a fundraising plan known as the "Countries Plan": holding auctions among the United States and other major governments to sell the rights to OpenAI's future AGI technology.

Brockman and Sutskever's logic was this: AGI has enormous national security significance and would sooner or later be taken from private companies by governments. Since it would have to be handed over eventually, why not sell it now in exchange for development funding?

"It sounded like tech people pitching a crazy startup, with an eccentric plan whose final step was: we'll get the government to pay for it," one person who heard the pitch recalled.

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

In December 2017, he wrote an email to Brockman, Sutskever, and other executives with blunt wording:

"In my view, this plan would not reduce, but rather intensify an arms race, potentially triggering human rights atrocities... and would undermine OpenAI's moral legitimacy." He insisted that OpenAI should optimize for "benefiting humanity."

An OpenAI spokesperson later told that the company "briefly considered whether to transform OpenAI into an international cooperation project," but "never considered selling AGI to any country."

The plan was ultimately shelved because of Microsoft's involvement. But Dario did not forget that Brockman and Sutskever had seriously discussed an idea he considered "morally repugnant."

Trust draining away bit by bit: hidden lists, rejected acquisition clauses, and the Kanye incident

Over the following two years, the rift continued to widen. Amodei began keeping records of moments when he believed Altman had misled him.

For example, during the private beta of GPT-3, Amodei asked Altman for a list of approved customers. Altman sent him one. But Amodei later claimed that he discovered another, longer list of approved customers, about which Altman had said nothing.

Another example: during negotiations with Microsoft over a $1 billion investment, Amodei strongly insisted on preserving the "merger and assistance clause" in OpenAI's charter — namely, that if a safer AGI project emerged, OpenAI should stop competing and assist it. He insisted that Microsoft had no right to veto a possible future acquisition of OpenAI, which was the most likely way to realize merger and assistance.

Altman told him Microsoft did not agree to this clause. But Amodei firmly believed that Altman had conceded — or simply never raised it — in order to rush the deal through, and then lied to him about it.

There were also some seemingly absurd business decisions that Amodei could not tolerate. In 2020, Altman met with rapper Kanye West and proposed letting him participate in the GPT-3 beta. At the time, Kanye had just released a video that appeared to show him urinating on a Grammy Award, and his mental state was highly unstable. Policy director Jack Clark ultimately stepped in and killed the dangerous idea.

Taken individually, these things might not have been enough to make someone resign. But in Amodei's mind, together they formed one conclusion: a technology as powerful as AGI must never be placed in the hands of such people.

The "plush toy advisory board," and an increasingly isolated safety faction

Amodei's sister, Daniela Amodei, had also grown weary of OpenAI.

As a manager who had worked in politics and possessed high emotional intelligence, Daniela became a complement to her brother's technical fervor. But as she took over more technical teams, her conflicts with management increased.

Interestingly, Daniela had a unique conflict-management mechanism — a plush toy advisory board.

She and board member Holden Karnofsky collected a large number of plush toys and assigned them roles: Jinji was a lazy cat who encouraged her to cancel meetings and do less work; Maura the pink bear was a detail-oriented Type A manager; and Beary Bonds the panda was the "guardian bear of generosity and empathy." She used this half-joking method to conduct management role-play.

By 2020, the Amodei siblings began building their own inner circle within OpenAI.

They set up private Slack channels filled with "safety jargon" that non-technical staff could not understand. A former colleague recalled: "Dario had a habit of simply saying: we must do X; it is the morally correct thing. It was like a shortcut to winning any argument — 'because safety.' This infuriated non-safety teams."

In 2019, Amodei at one point wanted to give up and even considered moving to DeepMind. But Jared Kaplan, a physics professor responsible for scaling law research, proposed a more disruptive idea: since their philosophies did not align, why not take everyone who cared about safety and start their own company?

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

Before the final showdown, Amodei presented Altman with a series of conditions for staying, including giving the safety team a "veto" over new product releases. He knew this would most likely be rejected, and only hoped to use it to force Altman to compromise.

During the negotiations, these seven people set up a secret Slack channel called "BATNA" — a business school term meaning "Best Alternative to a Negotiated Agreement," signaling they were ready to walk away at any time.

Unplugging the internet, the screen going dark

After confirming that Altman would not meet their demands, the "BATNA" team hired lawyers and began negotiating terms that came to be called a "divorce."

The two sides reached an agreement: the departing team could poach the employees it wanted within a few weeks, in exchange for a no-poaching commitment over the following year; they could take some internal research documents; and both sides would sign a non-disparagement agreement, never speaking ill of the other in public.

In December 2020, the fateful OpenAI all-hands online meeting was held.

Altman opened as usual, reporting on research progress. Then he handed the floor to Amodei.

Amodei calmly announced that he and six other leaders would resign and establish an AI organization focused on safety. The meeting chat window was instantly flooded with shocked emojis and exclamation marks. Everyone was guessing: who will train GPT-4? Who will be my new manager?

Amid the screen full of astonishment, Jack Clark typed a final line into the chat box: "I love you all."

Immediately after, in an instant, the screens of these seven people all went black.

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

A few hours later, this newly unemployed group of founders held their first video meeting. They had no computers, no GPUs, no data, and had not even settled on a company domain name — originally they wanted to call it Sparrow Systems, but ultimately changed it to Anthropic. Yet they raised glasses in celebration in front of the screen, laughing more happily than they had in months. They were free.

Trillion-dollar valuation and resentment that cannot be stopped

Five years later, reality and original intentions played a huge joke on each other.

In the founding vision Amodei wrote in 2021, he imagined a small AI safety research lab of about 50 people, and predicted that building AGI would cost only about $10 billion.

In the summer of 2023, the place still looked like the headquarters of a thriving cult or a university plant laboratory. One hundred and sixty safety-obsessed employees crowded together eating vegan sandwiches, stacking empty Liquid Death cans into a tall throne.

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

Its staff had ballooned to thousands, and it had moved into a high-rise building with heavy security. A dozen burly bodyguards stood at the entrance, and the floor where new versions of Claude were trained required special permissions even for internal employees.

Amodei had become a tech tycoon traveling by private jet and surrounded by bodyguards. Once low-key and reserved, he now had to manage one of the fastest-growing capital beasts in history, sprinting toward an IPO at a trillion-dollar valuation.

Daniela Amodei called all of this a "disaster of success": "It is stressful in every dimension. It is frightening, because we could lose it at any moment. That is only the business side. Then there is our mission, and we wonder, could we accidentally make things worse? For example, we think we understand everything about safety, but do we really?"

The only thing that has not changed is their mortal-enemy relationship with OpenAI.

According to people familiar with them, Dario and Daniela still despise Altman and Brockman. In Anthropic's internal Slack groups, Dario even said directly that Altman and OpenAI were "evil," so much so that during the 2023 OpenAI board crisis, public relations executives even asked him to delete the message for fear it would leak.

On OpenAI's side, employees privately described the Amodei siblings as "power-hungry virtue signalers," accusing them of hypocrisy — publicly warning about AI risks while privately accelerating the race.

This has become Silicon Valley's deepest black humor at the moment: both groups are pouring enormous computing power into training AI models to be kind, empathetic, and restrained; but as the humans who created them, they still cannot let go of their deep hatred for one another.