The Atlantic published an excerpt Tuesday from Kevin Roose’s forthcoming book The AGI Chronicles under the headline “OpenAI vs. Anthropic: Inside the Biggest Rivalry in Tech.” In a post on X, Roose described it as the story of “the feud between Dario Amodei and Sam Altman” and said Amodei left OpenAI with six colleagues in 2020 to start Anthropic.
The excerpt portrays the OpenAI-Anthropic split as rooted not just in business competition, but in a deeper clash over trust, safety and commercialization inside OpenAI as the company moved toward ever larger language models.
Roose’s account of the break
In the excerpt, Roose writes that Amodei had become disillusioned with OpenAI’s leadership by 2020, after helping lead work on GPT-2 and GPT-3. Roose reports that Amodei came to see Altman as “a two-faced manipulator” and OpenAI co-founder and president Greg Brockman as “power-hungry and reckless,” while also objecting to OpenAI’s conversion into a commercial enterprise and believing it was drifting from its original nonprofit mission.
Roose frames those tensions as central to the eventual formation of Anthropic. In his X post, he said the excerpt explains why Amodei and six colleagues left OpenAI in 2020 to found the rival company.
The “countries plan” dispute
One of the most striking episodes in the excerpt is an internal proposal Roose says dated back to 2017. He writes that Brockman and Ilya Sutskever considered what became known internally as the “countries plan,” which, according to five people familiar with it, involved an auction among governments including the United States, China and Russia for rights to OpenAI’s future AGI technology.
Roose writes that the idea began as a more benign concept for an international AGI coalition, but later evolved into a fundraising strategy. He also reports that Brockman and Sutskever justified the idea on the grounds that any future AGI would carry such major national-security implications that governments might seize it anyway.
OpenAI pushed back on that characterization. Roose writes that an OpenAI spokesperson said the company had “briefly considered whether to turn OpenAI into an international cooperative endeavor,” but “did not consider selling AGI to China or Russia.”
In the excerpt, Amodei is described as being appalled by the idea and warning internally that it would intensify arms races, risk human-rights abuses and undermine OpenAI’s moral legitimacy.
A Kanye West beta-access episode
The excerpt also includes a smaller but vivid example of the kinds of judgment calls that, in Roose’s telling, deepened Amodei’s mistrust of Altman. Roose writes that Altman offered Kanye West access to the GPT-3 private beta in 2020, citing two people familiar with the offer.
According to the excerpt, Jack Clark, then OpenAI’s policy director, stopped the idea.
Taken together, the episodes in the Atlantic excerpt sketch a version of the OpenAI-Anthropic rupture in which strategic disagreement and personal animosity were tightly intertwined, helping turn an internal fight over how to build advanced AI into one of the industry’s defining rivalries.
