Ex-OpenAI CTO Mira Murati Testifies Sam Altman Lied About AI Safety Reviews

The most damaging testimony yet in the Musk v. Altman trial came from someone who used to run OpenAI. Mira Murati — OpenAI's former chief technology officer who briefly served as interim CEO during the 2023 board crisis — testified in a recorded deposition that CEO Sam Altman was deceptive, told contradictory things to different executives, and created an atmosphere of distrust at the top of the company.

"My concern was about Sam saying one thing to one person and completely the opposite to another person," Murati said in the deposition, played Tuesday in Oakland federal court, per Reuters. She characterized Altman's conduct as "creating chaos."

Murati — who has since left OpenAI to co‑found her own AI startup — didn't hold back. When asked directly whether Altman had been truthful with her about a critical safety matter, she answered with a simple "No," The Verge reports.

The most specific allegation centers on a dispute over safety governance. Murati testified that Altman told her OpenAI's legal department had determined a new AI model did not need to go through the company's deployment safety board before release. When Murati checked with Jason Kwon — OpenAI's general counsel who is now chief strategy officer — his account didn't match.

"I confirmed that what Jason was saying and what Sam was saying were not the same thing," Murati testified, according to The Verge. She described it as a "misalignment" — corporate speak for one of them wasn't telling the truth.

When asked whether Altman was telling the truth, Murati answered: "No." This goes to the heart of Musk's case: that OpenAI, under Altman, prioritized speed over safety and that internal governance mechanisms were bypassed.

Murati painted a picture of leadership dysfunction that went beyond the safety review dispute. Altman undermined her ability to do her job, she said, by creating confusion about who had authority over what. "I had an incredibly hard job to do in an organization that was very complex," Murati testified. "I was asking Sam to lead, and lead with clarity, and not undermine my ability to do my job," The Verge reports.

When Altman was temporarily ousted by the board in 2023, Murati stepped in as interim CEO. She described the company as being at "catastrophic risk of falling apart" and said she feared "the company completely blowing up," per Reuters.

The contradiction — she didn't trust Altman but didn't want him gone — captures how deep the dysfunction ran: the company's top technical leader felt she couldn't trust the CEO's words, but also believed removing him would destroy the company.

Ars Technica notes that if Musk wins, OpenAI's for‑profit expansion could be blocked. If OpenAI wins, critics fear the nonprofit mission becomes unenforceable — Ars compares the concern to Google abandoning its "Don't be evil" motto.

The trial is structured in two phases: liability (jury advisory input) and remedies (judge‑only). Judge Yvonne Gonzalez Rogers makes the final decisions. The trial could last up to four weeks.

If you're a developer building on OpenAI's APIs, the trial raises real questions about company stability. A company whose CTO testifies the CEO lied about safety protocols is a company with unresolved leadership problems.

"The OpenAI trial could rewrite how AI companies are built," Yahoo Finance notes. A ruling forcing OpenAI to return to nonprofit status would fundamentally change its business model — and by extension, its API pricing, product roadmap, and enterprise contracts. Reputational damage could accelerate moves toward multi‑provider strategies.

Practical takeaway: don't build a business on a single model provider. AI companies are fragile organizations run by humans — and the humans running them don't always trust each other.

Beyond the personalities, this trial tests a fundamental question: can you start a nonprofit AI lab, take charitable donations, recruit with a public‑benefit mission, and then pivot to a for‑profit structure when the technology becomes commercially valuable?

Musk framed the stakes dramatically on X: "If I lose, it means it's OK to loot a charity," which would undermine "all charitable giving in the United States forever," Ars Technica reports. OpenAI posted on X that the lawsuit "has always been a baseless and jealous bid to derail a competitor."

Internal documents — including a 2017 diary entry from Greg Brockman reading "This is the only chance we have to get out from Elon," reported by Ars Technica — will shape how the industry thinks about AI governance for years. The era of AI labs as scrappy nonprofits with handshake governance is over. What replaces it is being decided in Oakland right now.