Dario Amodei’s AI safety contingent was growing disquieted with some of Sam Altman’s behaviors. Shortly after OpenAI’s Microsoft deal was inked in 2019, several of them were stunned to discover the extent of the promises that Altman had made to Microsoft for which technologies it would get access to in return for its investment. The terms of the deal didn’t align with what they had understood from Altman. If AI safety issues actually arose in OpenAI’s models, they worried, those commitments would make it far more difficult, if not impossible, to prevent the models’ deployment. Amodei’s contingent began to have serious doubts about Altman’s honesty.
“We’re all pragmatic people,” a person in the group says. “We’re obviously raising money; we’re going to do commercial stuff. It might look very reasonable if you’re someone who makes loads of deals like Sam, to be like, ‘All right, let’s make a deal, let’s trade a thing, we’re going to trade the next thing.’ And then if you are someone like me, you’re like, ‘We’re trading a thing we don’t fully understand.’ It feels like it commits us to an uncomfortable place.”
This was against the backdrop of a growing paranoia over different issues across the company. Within the AI safety contingent, it centered on what they saw as strengthening evidence that powerful misaligned systems could lead to disastrous outcomes. One bizarre experience in particular had left several of them somewhat nervous. In 2019, on a model trained after GPT‑2 with roughly twice the number of parameters, a group of researchers had begun advancing the AI safety work that Amodei had wanted: testing reinforcement learning from human feedback (RLHF) as a way to guide the model toward generating cheerful and positive content and away from anything offensive.
But late one night, a researcher made an update that included a single typo in his code before leaving the RLHF process to run overnight. That typo was an important one: It was a minus sign flipped to a plus sign that made the RLHF process work in reverse, pushing GPT‑2 to generate more offensive content instead of less. By the next morning, the typo had wreaked its havoc, and GPT‑2 was completing every single prompt with extremely lewd and sexually explicit language. It was hilarious—and also concerning. After identifying the error, the researcher pushed a fix to OpenAI’s code base with a comment: Let’s not make a utility minimizer.
In part fueled by the realization that scaling alone could produce more AI advancements, many employees also worried about what would happen if different companies caught on to OpenAI’s secret. “The secret of how our stuff works can be written on a grain of rice,” they would say to each other, meaning the single word scale. For the same reason, they worried about powerful capabilities landing in the hands of bad actors. Leadership leaned into this fear, frequently raising the threat of China, Russia, and North Korea and emphasizing the need for AGI development to stay in the hands of a US organization. At times this rankled employees who were not American. During lunches, they would question, Why did it have to be a US organization? remembers a former employee. Why not one from Europe? Why not one from China?
During these heady discussions philosophizing about the long‑term implications of AI research, many employees returned often to Altman’s early analogies between OpenAI and the Manhattan Project. Was OpenAI really building the equivalent of a nuclear weapon? It was a strange contrast to the plucky, idealistic culture it had built thus far as a largely academic organization. On Fridays, employees would kick back after a long week for music and wine nights, unwinding to the soothing sounds of a rotating cast of colleagues playing the office piano late into the night.
The shift in gravity unsettled some people, heightening their anxiety about random and unrelated incidents. Once, a journalist tailgated someone inside the gated parking lot to gain access to the building. Another time, an employee found an unaccounted‑for USB stick, stirring consternation about whether it contained malware files, a common vector of attack, and was some kind of attempt at a cybersecurity breach. After it was examined on an air‑gapped computer, one completely severed from the internet, the USB turned out to be nothing. At least twice, Amodei also used an air‑gapped computer to write critical strategy documents, connecting the machine directly to a printer to circulate only physical copies. He was paranoid about state actors stealing OpenAI’s secrets and building their own powerful AI models for malicious purposes.
“No one was prepared for this responsibility,” one employee remembers. “It kept people up at night.”
Altman himself was paranoid about people leaking information. He privately worried about Neuralink staff, with whom OpenAI continued to share an office, now with more unease after Elon Musk’s departure. Altman worried, too, about Musk, who wielded an extensive security apparatus including personal drivers and bodyguards. Keenly aware of the capability difference, Altman at one point secretly commissioned an electronic countersurveillance audit in an attempt to scan the office for any bugs that Musk may have left to spy on OpenAI.
To employees, Altman used the specter of US adversaries advancing AI research faster than OpenAI to rationalize why the company needed to be less and less open while working as fast as possible. “We must hold ourselves responsible for a good outcome for the world,” he wrote in his vision document. “On the other hand, if an authoritarian government builds AGI before we do and misuses it, we will have also failed at our mission—we almost certainly have to make rapid technical progress in order to succeed at our mission.”
In the author's note at the beginning of the book, Karen Hao notes, “I reached out to all of the key figures and companies that are described in this book to seek interviews and comment. OpenAI and Sam Altman chose not to cooperate.” Hao also attempted to reach Elon Musk for comment and got no response.
Excerpt adapted from Empire of AI: Dreams and Nightmares in Sam Altman’s OpenAI, by Karen Hao. Published by arrangement with Penguin Press. Copyright © 2025 by Karen Hao.
