Murati's Sworn Testimony That Altman Lied on Safety Could Redefine OpenAI's Legal Exposure

6. Murati's Sworn Testimony That Altman Lied on Safety Could Redefine OpenAI's Legal Exposure

Mira Murati, OpenAI's former CTO, testified under oath in the ongoing Musk v. Altman trial on Wednesday, April 30, that CEO Sam Altman lied to her about safety standards for a new AI model. In a video deposition shown to the court, Murati stated Altman falsely told her that OpenAI's legal department had cleared the model on safety grounds. The claim is not a leak or an anonymous source. It is sworn testimony from the company's former second-ranking technical officer.

That distinction matters enormously for OpenAI's regulatory and competitive position. The FTC has been probing OpenAI since 2023. The EU AI Act's high-risk classification framework places serious weight on documented internal safety processes. If Altman's representations to senior technical leadership were inaccurate, that is not just a governance failure inside one company. It is the kind of evidence that regulators in Washington and Brussels can use to argue that voluntary safety commitments from frontier labs are structurally unverifiable. Anthropic and Google DeepMind, both of which have built public identities around safety process transparency, stand to gain credibility by contrast. Neither needs to say a word.

The broader pattern worth watching: this trial is generating a factual record about OpenAI's internal decision-making that no press release or policy document can override. Elon Musk's legal team has every incentive to surface additional depositions from former insiders. OpenAI has at least three former senior figures, including Ilya Sutskever, whose accounts of the November 2023 board crisis have never been tested under oath. Each new deposition is a potential data point for regulators deciding how much to trust self-reported safety documentation from frontier AI developers.

Source: Mira Murati tells the court that she couldn't trust Sam Altman's words