Steven Bartlett, host of The Diary of a CEO, asked Karen Hao why the companies racing to build AGI are doing it. AGI, or artificial general intelligence, refers to systems meant to match people across most of what they can do. Bartlett noted that Hao had interviewed around 300 people, 80 or 90 of them from OpenAI, the maker of ChatGPT. Her answer was short. "I think it's because they're driven by an imperial agenda," she said. "And that is why I call these companies empires of AI."
Hao is a journalist who covers AI's effects on society. The phrase gave its name to her book, Empire of AI, an investigation of OpenAI that its publisher dates to May 20, 2025. The exchange aired on October 23, 2026, as a "Most Replayed Moment," a clip taken from an earlier full-length conversation.
What makes an empire
"Empire is the only metaphor that I've ever found to fully encapsulate all of the dimensions of what these companies do and the scale that they operate and what motivates them to do what they do," Hao said. She then set out the parallels she sees with "the empires of old."
The first is taking resources that belong to others. Training an AI model means feeding it huge amounts of material it can learn patterns from. Hao said the companies claim "the data of individuals, the intellectual property of artists, writers, and creators" for that purpose. She said they are also "land grabbing" to build the supercomputer facilities where the next generation of models is trained.
The second is labor. Hao said the companies contract "hundreds of thousands of workers all around the world, including in the U.S." to make these technologies. She gave no further breakdown in the clip. The companies also design their tools to automate labor, she argued, so the products erode labor rights once they are deployed. "This is a political choice that they have," she said. The publisher describes her book as tracing these material demands, from high-end chips and training data to data-cleaning labor, electricity and water, and it draws on the perspectives of Kenyan data workers and Chilean water activists.
The third trait is control over knowledge. According to Hao, the companies present themselves as the only ones who really understand the technology. Their message is that if the public dislikes AI, it does not know enough about it. She said they make that argument to the public and to policymakers, and that they have "captured the majority of the scientists" who study what AI can and cannot do.
The good empire and the evil one
Later in the clip, Hao added another trait she considers especially important for AI companies. Empires, she said, always tell the public: "we're the good empire. And we need to be an empire in the first place because there are also bad empires in the world." The promise is that if the public lets them take the resources and the labor, they will deliver progress and modernity for everyone, "this utopic state akin to an AI heaven." But "if the evil empire does it first, we will descend into a hell."
Today, she said, the evil empire is "most often" China. In OpenAI's early days it was Google. Hao described the company's reasoning then as a race against "this evil corporation that's driven by profit," with OpenAI cast as "a benevolent nonprofit."
The present-day version of that framing has a published counterpart in the writing of Anthropic CEO Dario Amodei. In Machines of Loving Grace, his October 2024 essay, Amodei argues that AI does not automatically favor democracy because it can also strengthen authoritarian surveillance and propaganda. He proposes a coalition of democracies that would secure an AI advantage through supply chains and chip restrictions while offering wider access to the benefits.
Two sides of the same coin
Bartlett asked whether the people building these companies really believe the outcome will be good for everyone, an "age of abundance." Hao said the industry's mythology depends on the belief that AI "could go very badly." "They need that part of the myth in order to then say, and that's why we need to be in control of the technology because that's the only way that it's going to go really, really well," she said.
She paraphrased OpenAI CEO Sam Altman as saying publicly that the worst case is "lights out for everyone," while the best case is curing cancer, solving climate change and abundance. She said Amodei uses "the same kind of rhetoric," with catastrophic or existential harm at one end and "mass human flourishing" at the other. "So this is like two sides of the same coin," she said. Together, she argued, the two stories justify "an extremely anti-democratic approach to AI development where there should not be broad participation in developing this technology. They must be the ones controlling it at every step of the way."
Altman's optimistic side is laid out in The Intelligence Age, his essay of September 23, 2024. It forecasts personal AI tutors, better healthcare, faster scientific discovery and climate solutions. It also ties that future to resources: Altman writes that broad access requires abundant chips, energy and affordable computing, and warns that without them AI could stay concentrated among the wealthy.
Hao's objection to keeping development in a few hands echoes another critic. Journalist Garrison Lovely, speaking on The Cognitive Revolution in September 2026, called the push to build machines that replace human labor risky and undemocratic. He said AGI should be built only with strong public buy-in.
A warning quote with two dates
The clip opened with an example of the double-edged rhetoric Hao described. Bartlett said he likes to track how people's quotes "evolve over time with their incentives." He then read a statement he attributed to Amodei in 2017, when Amodei was still at OpenAI. In it, Amodei describes "the Nick Bostrom style of fear that an AGI could destroy humanity" (Bostrom is a philosopher known for his work on AI risk) and puts his chance of a civilization-scale catastrophe at between 10% and 25%. Bartlett also read a 2019 line from Ilya Sutskever, whom Hao interviewed that year for a profile of OpenAI. Sutskever compared the future relationship between humans and autonomous AI to the way people build a highway between two cities without asking the animals for permission.
The Amodei quote appears to combine two different interviews. In a July 2017 interview with 80,000 Hours, Amodei treated Bostrom-style human extinction as a possible extreme outcome that deserved attention, while disagreeing that catastrophe was the default or nearly impossible to prevent. The 10% to 25% figure appears in a written summary of an October 2023 interview on The Logan Bartlett Show, a different podcast, by which time Amodei was Anthropic's CEO. That summary pairs his disaster estimate with a 75% to 90% chance that the technology goes well.