A sweeping New Yorker investigation revives old questions about OpenAI's CEO — and forces a harder one: whether any single executive should wield this much influence over a technology that reshapes everything.
In the fall of 2023, OpenAI's chief scientist Ilya Sutskever compiled roughly seventy pages of Slack messages, HR documents, and explanatory text into a set of secret memos. He sent them as disappearing messages to fellow board members. The subject was whether Sam Altman could be trusted to run the company. Sutskever told fellow board members at the time, a new investigation by The New Yorker reveals, "I don't think Sam is the guy who should have his finger on the button."
Sutskever's effort, of course, led to Altman's brief firing and swift reinstatement in November 2023 — an episode that ended with the board's dissolution and Altman's grip on OpenAI tighter than ever. Nearly two and a half years later, The New Yorker's Ronan Farrow and Andrew Marantz have surfaced new interviews and previously unseen documents that reopen the central question: not just whether Altman can be trusted, but whether the structure around him provides any meaningful check on his power.
The answer matters well beyond OpenAI's Slack channels. It matters for anyone who uses AI products, works in an industry being reshaped by them, or lives in a democracy trying to figure out how to govern them.
The Board That Couldn't Hold
The 2023 boardroom drama is well-documented. But the new reporting adds texture that shifts the story from corporate intrigue to governance failure. Sutskever was, by The New Yorker's account, "terrified" — taking cellphone photos of documents to avoid detection on company devices and using ephemeral messaging to communicate with sympathetic directors. The picture that emerges is of an insider who believed the stakes were existential but had almost no institutional support for acting on that belief.
OpenAI's unusual nonprofit structure was originally designed to protect humanity against rogue AI, as WIRED documented. It gave a small board with no financial stake the power to override commercial interests. In theory, this was a feature. In practice, when four directors tried to use that power, the backlash was immediate and overwhelming. Microsoft, employees, and investors rallied behind Altman. The board collapsed. The AI safety mechanism was dismantled by the very market forces it was supposed to counterbalance.
What replaced it was a more conventional board stacked with figures friendlier to Altman's vision. OpenAI has since moved toward a for-profit structure, evolving its organizational framework. The nonprofit shell remains, but the power dynamics have fundamentally shifted. The entity that was supposed to serve as a check now largely serves as a legacy artifact.
Altman's Theory of Power
Altman has not been shy about articulating his worldview, and it's worth taking seriously on its own terms. During an AMA session on X earlier this year, he laid out a framework that places democratic government above private companies in the hierarchy of legitimate authority. Altman wrote that he is "terrified of a world where AI companies act like they have more power than the government," Semafor reported.
That sounds reasonable. Altman's position is that a "democratically elected government" should take precedence over "unelected private companies." He has pledged that OpenAI would not conduct mass surveillance even if it became legal, calling it unconstitutional. He framed Anthropic's refusal to accept Pentagon contract terms as setting "an extremely scary precedent" — the precedent of a private company overriding government authority.
But Semafor's own analysis punctured the neatness of this framing. With trillion-dollar valuations and AI development outpacing regulators, the line between deferring to government and shaping government is blurry at best. When you're the CEO courting defense contracts, advising heads of state, and building technology that will mediate the flow of information for billions of people, "deference to government" can look a lot like partnership with government, on your terms.
As we reported in our coverage of the OpenAI-Anthropic Pentagon feud, the dispute centered on whether the Department of Defense could use AI systems for "any lawful use." Anthropic's CEO Dario Amodei demanded explicit prohibitions against domestic mass surveillance and autonomous weaponry. When Anthropic walked away, OpenAI stepped in and accepted similar terms while publicly claiming to maintain the same red lines. Amodei called that messaging "straight up lies."
The episode illustrates the core tension in Altman's philosophy. Deferring to government sounds principled until the government asks for something your competitor refused on ethical grounds — and you say yes.
Why This Isn't Just an OpenAI Story
The concentration of influence in a single AI executive is a sector-wide vulnerability, not a company-specific one. Altman is the most visible case, but the pattern repeats: charismatic founders with outsized personal authority, governance structures that look robust on paper but crumble under commercial pressure, and a regulatory environment that remains years behind the technology.
What makes the Altman case distinctive is the gap between OpenAI's founding rhetoric and its current reality. This was supposed to be the company that put safety above profit. Its nonprofit structure was supposed to ensure that. The board was supposed to be the failsafe. Every one of those safeguards has been weakened or removed, and the person who benefited most from their removal is the person they were designed to constrain.
The New Yorker's reporting doesn't resolve whether Altman is trustworthy — but that's almost beside the point. The real finding is that the question of trust shouldn't rest on any individual's character. Sutskever's seventy pages of documentation, his disappearing messages, his fear of detection — these are the behaviors of a whistleblower inside a system that offered no safe channel for dissent.
What Comes Next
Altman remains firmly in control of OpenAI. The company continues to expand into defense, enterprise, and consumer markets. Its for-profit transition is proceeding. No external regulator has imposed binding constraints on its operations.
The question isn't whether Altman will be removed again. It's whether anyone, such as a board, a regulator, or a competing institution, has the standing and the will to impose real accountability on the decisions being made inside OpenAI and companies like it. The 2023 board revolt showed that internal governance can't survive contact with billions of dollars in investor pressure. The Pentagon contract dispute showed that competitive dynamics push companies toward fewer restrictions, not more.
If there's a lesson in The New Yorker's investigation, it's this: the AI industry's governance problem is not a future risk. It already happened. The safeguards were tested, and they broke. What we're living with now is the aftermath. A period where the most consequential technology of the decade is being steered by executives whose accountability structures exist mostly on paper. Fixing that will require something more durable than a nonprofit charter or a charismatic founder's promise to do the right thing.
