Empire of AI: Dreams and Nightmares in Sam Altman's OpenAI

Altman is the central figure in Empire of AI. Altman is a controversial figure whose failings as a leader and technology entrepreneur are a focus of Empire of AI. Hao characterizes Altman as an untrustworthy manipulator who will say anything to achieve his goals.

Sam Altman grew up in an upper-middle-class family in St. Louis, Missouri. In 2005, he dropped out of Stanford University to work in Silicon Valley. He was part of the inaugural cohort of the Silicon Valley incubator Y Combinator (YC), and its cofounder, Paul Graham, became his mentor. Altman proved adept at building networks within the tech industry, and in 2014 he became president of YC. In 2015, Altman, along with Elon Musk, founded OpenAI. In 2019, Altman left his position as president of YC. Hao argues that Altman’s transition from YC to OpenAI is one of the earliest public examples of both his dishonesty and his use of public relations to cover up his failures and missteps. Altman publicly stated that he was going to step down from his role as president of YC and become chairman instead, thus making more time for OpenAI. She notes that, “The media widely reported Altman’s move as a well-choreographed step in his career and his new role as YC chairman. Except that he didn’t actually hold the title [of chairman],” he had simply said he did (69).

Altman’s mercurial personality governs leadership and decision-making at OpenAI. Hao documents Altman’s tendency tell every group what it wants to hear, sometimes resorting to dissimulation to do so. He will also outright lie to get what he wants. For instance, when he sensed that board member Toner was becoming skeptical of him and his role, he lied to another board member that Sutskever wanted her fired. Hao attributes this behavior to his “anxiety,” writing, that “his anxiety reach[ed] new heights and fuel[ed] his patterns of destructive behavior” (346). Employees and other leaders within OpenAI characterized his tendency to “[agree] with everyone to their face and […] [badmouth] them behind their backs” (346) as a form of “abuse and manipulation” (361). Hao reports that Altman has also engaged in morally questionable if not illegal behavior, as when he did not disclose his sole ownership of the OpenAI Startup Fund to the board.

Hao also documents how Altman’s relationship with his sister Annie has come under scrutiny after her public claims that he sexually abused her when they were children and did not help her financially, despite his vast wealth, when she was unhoused. In response, Altman began “to tell people that his sister had borderline personality disorder” (337) to discredit her claims. Hao compares this behavior to Altman’s other self-serving lies and dissimulations documented throughout Empire of AI.

Although Altman might be interpersonally toxic, he has cultivated a loyal following among his employees and some tech leaders, like Microsoft CEO Satya Nadella, as demonstrated by their dedicated support for him and his vision during the attempted board coup in 2023.

OpenAI is the technology development organization that is a key focus of Empire of AI, although Hao also reports on the actions of other corporations in the field, including traditional big tech firms like Google and Microsoft. Hao describes OpenAI’s history and corporate structure as distinct from traditional tech firms like Microsoft.

OpenAI was founded by Altman and Elon Musk in 2015 as a non-profit entity. They saw it as a “Manhattan Project for AI” (26). Just as the Manhattan Project sought to develop nuclear weapons before the Axis Powers during World War II, OpenAI wanted to develop a potentially life-ending technology, AGI, to ensure it was “aligned” with humanity’s goals. They decided to found the entity as a non-profit to ensure that it would be used “for the good of the world” (27). The name “Open” in the title was broadly interpreted as an indication that OpenAI’s research would be made available to the public, but the founders internally acknowledged that this was not the case.

Under the guidance of Ilya Sutskever, OpenAI became focused on the need to increase processing power in order to realize its goal of AGI. To do so, OpenAI needed to buy more GPUs. To raise the funds for the enormous capital outlay, Altman created “a limited partnership, or LP, to act as a for-profit arm for receiving investment and commercializing OpenAI’s technologies” (66). Altman framed this as a way of ensuring that OpenAI could achieve its mission, but the organization became increasingly focused on releasing commercial products. Altman became CEO of the LP, and many employees resigned from the nonprofit and joined the LP.

In 2024, OpenAI raised $6.6 billion in funding from investors. However, there was a “hitch,” “investors could demand their money back if the company did not convert to a for-profit in two years” (405). OpenAI later announced a new corporate structure as “a for-profit public benefit corporation” with a nonprofit that had “equity with shares in the for-profit” (406). This is a highly unique corporate structure. Hao suggests that this transition from nonprofit to for-profit “could set a dangerous precedent for many more startups to designate themselves as nonprofits,” thus securing tax write-offs until they have developed a product (406).

Greg Brockman is a cofounder of OpenAI. Brockman, an engineer who grew up in North Dakota, was an early investor and chief technology officer for the payment processor Stripe before joining OpenAI. Hao describes Brockman as “tall and stocky with an amiable demeanor” (46). In an interview, Brockman tells Hao that his leadership style is based on the importance of “cohesion” and “being physically together [to] help[] with the serendipitous exchange of ideas” (54). Hao documents complaints about Brockman’s leadership at OpenAI. She notes, for example, that OpenAI chief technology officer Mira Murati felt that “Brockman was like a second CEO but a bad one—highly opinionated and prone to driving people to burnout with his intensity” (346). He would dive into projects without coordinating with other members of leadership or employees. Hao reports that Brockman was highly resistant to oversight and refused to listen to anyone but Altman. This created tension and stress for employees and exacerbated inter-company conflicts. Murati also felt that Brockman “enabled” Altman’s “dumb ideas” (350).

Ilya Sutskever was the chief scientist of OpenAI from 2015 to 2024, when he left to create his own AI startup. A math prodigy who studied computer science at the University of Toronto, Sutskever is recognized as a leading developer of deep learning neural networks like that used by ChatGPT.

Hao characterizes Sutskever as a true believer in the utopian and apocalyptic possibilities of AGI. She describes him as “OpenAI’s own resident mystic” (254). He is highly driven to develop advanced AI technologies, and he believes “AGI [is] imminent” (254). He is also concerned about the potential for an AI apocalypse. He left OpenAI in part because he felt they were no longer taking “safety” concerns seriously.

As described by Hao, Sutskever was a key figure in the attempted coup of Sam Altman from the position of CEO by the board. Sutskever had come to believe that Altman had engaged in a pattern of abuse (353) and had acted recklessly by not ensuring products were “safe” before releasing them to the public. However, when public and employee support fell behind Altman, Sutskever relented and agreed to remain with the company under Altman’s leadership, at least temporarily.