Jurors hear testimony that OpenAI’s nonprofit board failed to exercise meaningful oversight

The fight in Oakland is not really about one memo or one product launch. It is about whether OpenAI’s unusual structure ever worked the way it was supposed to. In Elon Musk’s lawsuit, the core claim is that OpenAI’s nonprofit parent was meant to control the company in service of humanity, but the board stopped functioning as real oversight while the for-profit side got bigger, richer, and faster. On Thursday, jurors heard some of the clearest testimony yet on that point — from former board members Helen Toner and Tasha McCauley, plus former safety staffer Rosie Campbell. (nbcbayarea.com) ### What did jurors actually hear? They heard edited video depositions from Toner and McCauley rather than live testimony, and both described a board that struggled to get reliable information from Sam Altman. McCauley said the nonprofit board did not have enough confidence that it was fully informed to make decisions, which matters because OpenAI’s design gave that board the job of guarding the mission, not just chasing returns. (nbcbayarea.com) ### Why does the nonprofit board matter so much? Because OpenAI was built with a weird promise. The nonprofit parent was supposed to sit on top of the whole structure and make sure AGI was developed for humanity’s benefit. The for-profit arm could raise money and build products, but in theory it was still subordinate to the mission. Musk’s case rises or falls on whether that arrangement became mostly decorative. TechCrunch’s read is basically that the lawsuit hinges on whether the for-profit subsidiary advanced that mission or undermined it. (techcrunch.com) ### What was the most damaging detail? McCauley pointed to a safety-review issue involving ChatGPT variants. She said Altman told the board that three variants had gone through the company’s deployment safety board, when only one had at that point. She described a broader pattern of dishonesty and said that kind of gap made meaningful oversight hard, because a board cannot supervise what it cannot accurately see. (courthousenews.com) ### Where does safety come into this? Rosie Campbell connected the governance problem to product pressure. She testified that OpenAI felt more research-focused when she joined in 2021, but became more product-focused over time. She said OpenAI disbanded her AGI readiness team by the end of 2024, and another safety-focused group, the Superalignment team, was also shut down around that period. (courthousenews.com) ### Was there a concrete example of process breaking down? Yes — Campbell pointed to a GPT-4 deployment in India through Microsoft’s Bing before OpenAI’s Deployment Safety Board had reviewed it. Her point was not that this specific release was catastrophic. It was that safety procedures only matter if everyone treats them as binding precedent before the models get much more powerful. That incident also surfaced as one of the warning signs tied to Altman’s brief ouster in November 2023. (techcrunch.com) ### Is this just Musk attacking Altman? Not entirely. Musk obviously has his own motives, and OpenAI has spent much of the trial trying to show that he supported for-profit options before leaving. Shivon Zilis testified that OpenAI’s structure was debated “ad nauseam,” including different for-profit paths, and that Musk at one point wanted OpenAI folded into Tesla. So the defense argument is: Musk is not some pure guardian of nonprofit virtue here. (cnbc.com) ### Then what is the jury really being asked to decide? Not whether commercialization happened — that part is obvious. The question is whether commercialization outran governance. If the nonprofit board could not get candid answers, could not enforce safety gates, and could not hold the CEO in check, then the structure that justified OpenAI’s hybrid model may have stopped being real in practice. That is why this testimony matters beyond Silicon Valley gossip. (courthousenews.com) ### Why does this matter outside the courtroom? Because a lot of AI governance talk assumes you can solve dangerous incentives with clever org charts. OpenAI has been the flagship example — nonprofit control on top, commercial engine underneath. But this week’s testimony suggests the hard part is not drawing the boxes. It is making sure the people in those boxes can actually say no. (courthousenews.com) ### Bottom line? The testimony jurors heard was less about one dramatic betrayal than a slower failure: the board that was supposed to restrain OpenAI’s commercial machine may not have had the information, leverage, or internal trust to do the job. If that lands with the jury — and later with the judge — the case becomes a referendum on whether mission-first AI governance can survive founder power and hyperscale money. (nbcbayarea.com)