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

Hao frames this as a book about Sam Altman and OpenAI. What it actually is — the version worth reading — is a book about how an institution built around a stated mission can be hollowed out while the mission branding intensifies. The AI safety story is secondary. The organizational capture story is not.

The Mission Drift

OpenAI was founded as a nonprofit in 2015 with the explicit goal of ensuring that artificial general intelligence benefits humanity — and with a structure designed to keep commercial pressure from distorting that goal. By 2019 it had a "capped-profit" subsidiary. By 2023 it was generating billions in revenue and had Microsoft as its largest investor. By 2025 it was restructuring into a fully for-profit entity.

Hao is good at tracing the ratchet. Each step had a justification — we need capital to compete, compute costs are enormous, talent requires equity. Each justification was true. What she shows is that the ratchet only moves in one direction. No amount of commercial success triggers a return toward the original structure. The mission doesn't constrain the institution; the institution redefines the mission.

The most clarifying moment in the book is the November 2023 board crisis. The board had formal authority to remove Altman — and exercised it. Within days, Microsoft signaled it would hire Altman and the entire research team. Hundreds of employees signed a letter threatening to resign. The board reinstated Altman and then itself resigned. The sequence reveals exactly where actual power resided, and it was not with the governance structure. Formal authority without operational leverage is theater.

The Altman Character Study

Hao had significant access, and the Altman portrait that emerges is careful and unflattering in ways that don't feel tendentious. He is genuinely mission-driven in the sense that the mission is real to him — and also genuinely skilled at making the mission serve his interests, in ways that may not be fully legible even to himself.

The particular talent Hao identifies: Altman makes people feel like insiders. Investors, governments, researchers, journalists — all get the sense they're in the room where decisions are made, that they understand the stakes, that their involvement matters. This is not simple flattery. It's a durable mechanism for building coalitions that are difficult to defect from, because defection means admitting you were not actually inside.

Where It Falls Short

The book is strongest as character study and weakest as structural analysis. Hao traces what happened inside OpenAI with precision and sourcing. She is less rigorous about why the broader pattern — safety-focused AI labs drifting toward commercial deployment under competitive pressure — was probably inevitable given the incentive structure, regardless of who sat in which chair.

The "nightmares" in the subtitle point toward AI risk scenarios that Hao gestures at but doesn't develop with the same depth as the organizational history. The result is a book that explains how OpenAI became what it became, without quite making the case for why that trajectory was locked in from the beginning. Those are different arguments, and the second one — the structural case — would have made the book harder to dismiss as a personality profile.

The governance failure at OpenAI matters beyond OpenAI. Every institution claiming to do dangerous work responsibly while generating enormous commercial returns faces a version of the same problem: the safety mission requires constraints, constraints reduce competitive position, competitive pressure erodes constraints. Hao documents one instance of this cycle clearly enough that the pattern is hard to ignore. Worth finishing.