OpenAI once considered a controversial plan involving the US, China and Russia for access to its future artificial general intelligence (AGI), according to reporting by The Atlantic. The proposal, which emerged while the company was seeking funding, reportedly alarmed senior researcher Dario Amodei and became one of several disputes that contributed to his eventual departure.
The same account, based on more than 150 interviews by journalist Kevin Roose for his forthcoming book The AGI Chronicles, describes another unusual episode: Sam Altman allegedly offered Kanye West access to the private beta of GPT-3 in 2020. OpenAI policy director Jack Clark reportedly intervened and stopped the proposal.
The revelations offer new detail about the tensions inside OpenAI before Amodei left and later co-founded Anthropic. They also come with an important qualification: OpenAI disputes the characterisation of the China-Russia proposal, saying it briefly considered an international cooperative structure but did not consider selling AGI to either country.
OpenAI's Disputed 'Countries Plan'
According to Roose's reporting, the origins of the so-called 'countries plan' went back to 2017, when OpenAI co-founders Greg Brockman and Ilya Sutskever were considering ways to secure funding.
Brockman and Sutskever reportedly proposed creating an international coalition that could share the benefits of AGI and manage its risks. But the idea allegedly developed into a proposal for governments, including the US, China and Russia, to compete for rights to OpenAI's future technology.
The reasoning, according to people familiar with the discussions, was that governments could eventually seize advanced AI technology because of its national-security importance. If that was inevitable, the argument went, OpenAI could potentially raise money by allowing governments to pay for access.
Amodei strongly objected. In a 2017 email Roose cited, he warned that the proposal could intensify an international arms race, enable human-rights abuses and damage OpenAI's legitimacy.
OpenAI has disputed the central allegation. A spokesperson told The Atlantic that the company had 'briefly considered whether to turn OpenAI into an international cooperative endeavour', but 'did not consider selling AGI to China or Russia'. That distinction is important because the report describes internal discussions and proposals rather than a completed agreement or transaction.
The Kanye West GPT-3 Episode
The report also describes a separate dispute involving Kanye West, who is now known as Ye.
Roose reports that Altman met West in 2020 and offered him access to the private beta of GPT-3. Two people familiar with the offer reportedly told Roose that Clark, then OpenAI's policy director, blocked the arrangement.
The episode mattered because it allegedly reinforced Amodei's concerns about how OpenAI was making decisions around access to increasingly powerful AI systems.
GPT-3 was particularly important at the time. OpenAI introduced the model in 2020 and subsequently expanded access through its API. The company described GPT-3 as one of its major research breakthroughs before ChatGPT transformed public awareness of generative AI.
Why Amodei Eventually Left OpenAI
Roose reported the disagreements were broader than the two episodes highlighted by the headline. His account describes Amodei becoming increasingly frustrated with OpenAI's commercial direction, leadership and approach to AI safety.
OpenAI officially announced Amodei's departure on December 29, 2020, saying he had spent nearly five years at the company and had helped lead work on GPT-2 and GPT-3. The company said Amodei and several colleagues were planning a new project focused more heavily on research.
That project became Anthropic, which Amodei co-founded with his sister Daniela and other former OpenAI researchers. The company has since developed into one of OpenAI's major competitors, with safety and AI risk remaining central themes of Amodei's public work.
The disagreements described by Roose helped shape the creation of Anthropic and, in turn, the competitive landscape of frontier AI.
Today, Amodei and Altman lead rival companies competing to develop increasingly capable systems while publicly debating how quickly that technology should advance. Recent reporting shows Amodei continuing to argue for stronger safeguards and international coordination as AI capabilities accelerate.
The new account does not establish that OpenAI ever reached deals with China or Russia. Instead, it provides a disputed reconstruction of internal ideas that, according to Roose, helped expose a fundamental disagreement over how a powerful AI company should balance funding, commercial expansion, national security and safety.
