On Monday a jury in Oakland cleared Sam Altman and OpenAI in Elon Musk’s fraud case. It took less than two hours, and it decided one thing: Musk sued too late.
The court never ruled on whether Altman betrayed OpenAI’s founding mission. So the question the trial was meant to settle is still open, and it sits at the centre of a remarkable New Yorker investigation.
Ronan Farrow and Andrew Marantz spent eighteen months on the piece, speaking to more than a hundred people and reading internal memos. What they assembled was less a list of incidents than a pattern that runs back fifteen years.
The pattern is that Altman builds structures that constrain him, then takes them apart when the time comes to be constrained. The question is not whether it exists. It is why it keeps working.
A pattern that predates OpenAI
Paul Graham recruited Altman to run Y Combinator in 2014 and once wrote that, dropped onto an island of cannibals, he would return as king. By 2018 partners were complaining, and Graham asked him to step down.
The public version was an orderly handover to chairman. The private version was different. Graham later told colleagues that Altman had been lying to them the whole time.
In November 2023 the OpenAI board tried the same thing, armed with a memo from co-founder Ilya Sutskever that opened with the word “lying.” Altman learned over a video call that he was out.
He was back within five days. The board members who removed him were gone within weeks. An independent review let him stay, and the firing simply did not hold.
How the mechanism works
Former researcher Carroll Wainwright explained that Altman sets up structures that constrain him on paper, then does away with them when the future arrives. The structures often look like genuine commitments while they are being made.
During the 2019 Microsoft negotiation, Dario Amodei found a clause had been added letting Microsoft block OpenAI’s mergers, gutting a safety provision he had ranked first. He read the text aloud. Altman denied it existed, then said he could not remember.
The language does similar work. In 2017 Altman told US intelligence officials that China was running an “A.G.I. Manhattan Project” and OpenAI needed government billions to keep pace. Asked for evidence, he said he had heard things, and never followed up.
The fear and the fundraising arrived in the same sentence. The skill is genuinely rare, and the thing that makes Altman hard to constrain is the same thing that makes him effective.
What it has cost
OpenAI launched its Superalignment team in July 2023, pledging a fifth of its compute to the problem of controlling advanced AI. Staff told Farrow and Marantz the real allocation was closer to one or two per cent, much of it on the oldest chips.
Sutskever and Jan Leike resigned in May 2024 and the team dissolved within days. Asked to interview researchers on existential safety, a company representative replied that it was not, like, a thing.
The morning after a man threw a Molotov cocktail at his house in April, Altman published a blog post arguing that power cannot be too concentrated. Five weeks earlier he had told staff who questioned a Pentagon deal that they did not get to weigh in.
The verdict, and what it left untouched
This week’s ruling fits the longer story. The advisory jury found Musk’s claims fell outside the statute of limitations, and Judge Yvonne Gonzalez Rogers adopted that finding, noting substantial evidence behind it. Musk says he will appeal.
Whether Altman and Greg Brockman abandoned a charitable mission was never decided. The case turned on timing, and OpenAI now holds a valuation near $852 billion as it moves toward the public markets.
Marantz framed the real issue on a podcast afterwards. The point is not which founder should hold this kind of power. It is who is handing any of them that much of it.
The pattern across Y Combinator and OpenAI is documented. The dismantled safety teams and Amodei’s two hundred pages of notes are documented. What is left, once the valuations and the Tolkien metaphors are stripped away, is one man’s relationship to the truth, and a row of institutions that have struggled to test it.
