The Verge has reported that Mira Murati, OpenAI's former Chief Technology Officer, testified under oath during the ongoing Musk v. Altman trial that CEO Sam Altman misled her about the safety evaluation process for a new artificial intelligence model. In video deposition footage shown in court, Murati stated that Altman falsely claimed OpenAI's legal department had cleared a new model under a particular standard, a claim she says proved to be untrue.
The significance of this testimony extends well beyond the specific legal dispute in which it appeared. Murati was not a peripheral figure at OpenAI. She served as the company's technical conscience for years and, critically, stepped into the role of interim CEO during the chaotic days in November 2023 when the board briefly removed Altman before reversing course and reinstating him. Her willingness to offer sworn testimony describing Altman as someone whose words she could not trust is, by any measure, a striking development for a company that has positioned itself as the responsible steward of one of the most consequential technologies in modern history.
The broader context here involves a persistent tension at the heart of OpenAI's identity. The organization was founded on the premise that artificial intelligence safety was not merely a technical problem but a moral obligation, one serious enough to justify a nonprofit governance structure designed to resist the usual commercial pressures. That structure has been stress-tested repeatedly. The board's attempted ouster of Altman, the subsequent exodus of several senior researchers focused on safety work, and Murati's own departure from the company last year all form part of the same pattern. What emerges is a picture of an institution where the gap between its stated values and its internal decision-making has been a source of sustained friction.
Murati's specific allegation, as reported by The Verge, concerns the safety standards applied to a model release, which places it directly in the territory that critics of OpenAI have been most vocal about. A number of former employees, including some who helped found OpenAI or occupied senior roles, have raised concerns over time that competitive pressure, particularly from rivals moving quickly to release capable models, has shortened the runway given to safety evaluation. Whether or not those concerns are legally actionable, they speak to a cultural shift that observers of the AI industry have been tracking carefully. When a company's former second-in-command says under oath that she could not trust what the CEO told her about legal and safety clearances, it gives those concerns a very different weight than an anonymous complaint or a resigned statement.
The trial itself, initiated by Elon Musk, centers on different grievances related to OpenAI's transition away from its original nonprofit structure and toward a more conventional commercial model. But litigation has a way of surfacing testimony that travels far beyond the immediate dispute. Murati's deposition is a reminder that courtrooms, unlike press releases or carefully managed departures, operate under oath and without the usual filters of corporate communications. The likely reading is that this testimony will do lasting damage to Altman's credibility with at least some audiences, even if OpenAI ultimately prevails in the legal proceedings or if the specific allegations prove difficult to adjudicate.
The consequences extend to several different groups. For OpenAI's current employees, particularly those who work on safety and alignment, the deposition from someone they knew and respected amounts to a formal record of something many may have suspected. For regulators in Washington and Brussels who are already examining how frontier AI models are evaluated before release, sworn testimony about misleading internal communications around safety standards is precisely the kind of material that sharpens legislative interest. For the broader ecosystem of investors, partners, and enterprise customers who have committed significant resources to OpenAI's roadmap, the suggestion that internal governance may be less rigorous than advertised introduces a new category of reputational risk to evaluate.
There is also an indirect effect on the competitive landscape. OpenAI's strongest argument against those who would regulate AI development more aggressively has always been that it polices itself more carefully than a purely commercial actor would. Murati's testimony complicates that argument in ways that competitors, whether Anthropic with its explicit safety focus or Google DeepMind with its institutional resources, may find useful.
What to watch for next is whether other former senior figures are drawn into the proceedings in any capacity, either through depositions of their own or through the documentary record that trials tend to surface. Also worth monitoring is how OpenAI responds formally, since its public posture so far has generally been to minimize the significance of internal disputes. Perhaps most important is whether any regulatory body in a major jurisdiction treats this testimony as grounds for closer examination of the company's model evaluation practices. That would represent a significant escalation of external oversight, one that OpenAI has managed to avoid despite years of public controversy.
