Mira Murati, former CTO of OpenAI, testified that Sam Altman misled her regarding safety reviews for a new AI model. This testimony occurred during the ongoing Musk v. Altman trial, where Murati stated Altman falsely claimed the legal department had cleared the model for deployment without needing the safety board's approval.
The legal battle between Elon Musk and OpenAI CEO Sam Altman has brought forth testimony from Mira Murati, the company's former Chief Technology Officer. Murati stated under oath that Altman had not been truthful with her about the safety protocols for a new AI model. In a video deposition played in court, Murati recounted a conversation where Altman asserted that OpenAI's legal team had determined the model did not require review by the company's deployment safety board. When questioned if Altman was being truthful, Murati responded with a definitive "No."
Murati's testimony suggests a potential discrepancy between Altman's statements and the company's internal safety procedures. OpenAI has established a Deployment Safety Board (DSB) that jointly reviews AI models with Microsoft for deployment above certain capability thresholds. This board focuses specifically on deployment decisions rather than earlier stages of model development. Murati's account implies that Altman may have bypassed or misrepresented the necessity of this review process. She further testified that she investigated the matter by speaking with Jason Kwon, who is now the chief strategy officer at OpenAI. Murati noted an "misalignment" between Kwon and Altman and took steps to ensure the model underwent the safety board's review.
The trial, initiated by Elon Musk, centers on allegations that Altman and OpenAI betrayed the company's original nonprofit mission by shifting towards a for-profit structure. Musk claims this transition involved deception and a departure from the commitment to benefit humanity, rather than generating profit. OpenAI, in its defense, argues that Musk's lawsuit is an attempt to hinder its progress due to the success of his own competing AI company, xAI.
Murati's role in the trial is significant, given her former position as CTO and her brief tenure as interim CEO following Altman's temporary ouster in November 2023. Her testimony adds a layer of internal perspective on the leadership dynamics and operational practices at OpenAI. She also described experiencing managerial friction during her time at the company, stating that her criticisms were "completely management related" and that she found her role to be "an incredibly hard job" within a complex organization.
The Musk v. Altman trial has seen various testimonies shedding light on OpenAI's internal workings and governance. Greg Brockman, president of OpenAI, also testified, as did Shivon Zilis, a former OpenAI board member and mother of some of Musk's children. Elon Musk himself testified, detailing his loss of confidence in Altman's commitment to the nonprofit mission. He expressed concerns about Microsoft's substantial investment in OpenAI, questioning whether Microsoft should have control over advanced AI capabilities.
OpenAI's approach to safety has been a recurring theme. The company has detailed its safety practices, including a cross-functional Safety Advisory Group that reviews model capabilities and makes recommendations before deployment. Final decisions rest with company leadership, with the Board of Directors providing oversight. Furthermore, OpenAI has a joint Deployment Safety Board with Microsoft to approve the release of models above a certain capability threshold. However, reports suggest that in some instances, models like GPT-4 were tested publicly before receiving DSB approval. The company's Preparedness Framework outlines risk assessment processes, but questions have been raised about its effectiveness and the potential for leadership to override safety recommendations.
Murati's testimony regarding the safety review process for the new AI model could be critical in establishing whether Altman adhered to OpenAI's established protocols. Her account suggests a potential pattern of Altman prioritizing speed or other objectives over strict adherence to safety review procedures, a claim that aligns with some of the broader allegations in Musk's lawsuit regarding the company's shift away from its original mission. The trial is expected to continue, with further testimony potentially revealing more about the internal decision-making processes and the integrity of OpenAI's safety measures.