Murati Testifies Against Altman's Lies: OpenAI Safety Review Insider Details Exposed

OpenAI Former CTO Murati Testifies Under Oath: Altman Lied About Safety Reviews

OpenAI's former Chief Technology Officer Mira Murati testified under oath in court, accusing CEO Sam Altman of lying to her about AI model safety standards. This explosive testimony emerged during the ongoing Musk v. Altman trial, adding significant weight to the legal battle over OpenAI's governance and mission.

In video testimony played during Wednesday's hearing, Murati stated that Altman had falsely claimed OpenAI's legal department had determined that a new AI model did not require additional safety review procedures. In other words, Altman allegedly bypassed internal safety compliance processes and used the legal department's name to endorse this decision — when in fact no such endorsement existed.

Notably, within the U.S. legal system, testimony under oath carries extremely high legal weight. Making false statements after being sworn in constitutes perjury, punishable by up to 5 years of federal imprisonment. This means every word of Murati's testimony was rigorously vetted by both herself and her legal team, and her willingness to accept the legal risk of perjury charges speaks to the seriousness of these accusations. Video depositions are typically recorded before formal trial proceedings, with attorneys from both sides having the opportunity to conduct cross-examination, giving them evidentiary weight essentially equivalent to live courtroom testimony.

Trust Collapse: CTO Publicly States She "Cannot Trust the CEO's Words"

Murati's testimony reveals more than just a single specific lie — it exposes a deep rift within OpenAI on AI safety issues. As the former CTO, Murati was one of the core decision-makers on OpenAI's technical roadmap. Her public statement in court that she "cannot trust Altman's words" means that during her tenure, there was a severe trust breakdown between the company's top management and technical leadership regarding the enforcement of safety standards.

The weight of this testimony lies in its legal context — this is sworn testimony (under oath), and Murati bears legal responsibility for the truthfulness of her statements. This isn't a social media call-out from afar, but a carefully considered, legally binding formal accusation.

Safety Review Processes Were Bypassed

From Murati's testimony, it can be inferred that disputes existed within OpenAI regarding safety evaluation procedures before new model releases. Altman is accused of attempting to accelerate model development timelines, even resorting to fabricated legal opinions to circumvent safety reviews. This echoes longstanding external skepticism about whether OpenAI's "safety first" commitment truly lives up to its name.

To understand the severity of this accusation, one needs to understand the design intent behind OpenAI's safety review mechanisms. Before releasing new models, OpenAI typically requires multi-layered safety evaluation processes, including red teaming, alignment evaluation, and internal safety committee review. Red teaming involves dedicated teams simulating malicious users attempting to breach a model's safety restrictions to discover potential risks — for example, testing whether a model would provide guidance on manufacturing dangerous substances or could be induced to generate harmful content. Alignment evaluation examines whether a model's outputs conform to human values and safety guidelines, ensuring the model doesn't produce unintended harmful behaviors while pursuing its objectives. OpenAI also published a Preparedness Framework that explicitly defines release standards for models at different risk levels, categorizing risks into "low," "medium," "high," and "critical" tiers, requiring that only models with risk assessments below the "high" level can be publicly released. Murati's testimony suggests that these carefully designed safety processes may have been effectively neutralized by management through administrative means.

Musk v. Altman: Why Murati's Testimony Is Crucial

Murati's testimony is a critical piece of the Musk v. Altman case. Musk, as a co-founder of OpenAI and its primary early funder, accuses Altman of betraying OpenAI's founding mission as a nonprofit organization dedicated to benefiting all of humanity, instead pursuing the maximization of commercial interests.

This lawsuit has a complex historical background. Musk first filed suit against Altman and OpenAI in early 2024, later withdrew it, and refiled later that same year. The core dispute traces back to OpenAI's founding structure: when established in 2015, OpenAI was a 501(c)(3) nonprofit organization, and Musk cumulatively donated approximately $44 million. In 2019, OpenAI created a "capped-profit" for-profit subsidiary — investor returns were limited to 100 times their initial investment, with any excess going to the nonprofit parent. However, OpenAI subsequently pushed further toward converting into a fully for-profit company, planning to remove the profit cap and change the company's legal structure. Musk argues that this series of transformations violated the agreements under which he originally provided funding — namely, that OpenAI should develop AGI (Artificial General Intelligence) in an open, nonprofit manner for the benefit of all humanity. The case involves multiple legal claims including breach of contract, breach of fiduciary duty, and unjust enrichment.

Murati's testimony provides powerful support for Musk's accusations from an insider perspective: if the CEO is willing to bypass even internal safety processes through deception, how much credibility remains for the grander promises about company mission and governance structure?

What This Trial Means for the AI Industry

The significance of this trial extends far beyond a single company, touching on the most fundamental questions facing the AI industry:

• Balancing safety and speed: In the fierce AI race, are safety reviews being systematically weakened?
• Corporate governance transparency: Are AI companies' internal decision-making processes sufficiently transparent and accountable?
• Voice of technical leaders: When technical leaders and business decision-makers disagree, who should have the final say?

The urgency of these questions lies in the fact that AI model safety evaluations currently rely primarily on corporate self-regulation, lacking mandatory external audit mechanisms. The United States has not yet passed comprehensive federal AI safety legislation. While the Biden administration issued an AI Executive Order in 2023 requiring developers to report to the government when training models exceeding certain compute thresholds, this executive order has since been revoked by the subsequent administration. The EU's AI Act, while establishing a risk-based tiered regulatory framework that classifies AI systems into "unacceptable risk," "high risk," "limited risk," and "minimal risk" categories with varying degrees of regulatory requirements, is still developing specific safety evaluation standards for frontier general-purpose AI models. In the absence of external mandatory constraints, internal corporate safety review processes become the last line of defense — and Murati's testimony suggests that even this line of defense can be easily breached.

Murati's Departure and the Brain Drain from OpenAI's Safety Teams

Murati had previously left OpenAI. Her departure itself had triggered widespread speculation about internal conflicts at OpenAI, and today's court testimony largely confirms those suspicions — when a CTO discovers she cannot trust the CEO's statements on safety issues, leaving becomes almost inevitable.

Murati is not the only executive to depart. In recent years, multiple core members focused on AI safety have successively left OpenAI, forming a concerning trend. In May 2024, Ilya Sutskever and Jan Leike, co-leads of OpenAI's Superalignment team, departed in succession. The Superalignment team was established in 2023 with the mission of solving the fundamental challenge of ensuring that AI systems surpassing human intelligence remain controllable. Upon leaving, Leike publicly stated that "safety culture and processes have taken a backseat to shiny products," directly pointing out that the company's priorities between safety and commerce had undergone a fundamental shift. Previously, Sutskever had been one of the key drivers behind the OpenAI board's brief firing of Altman in November 2023 — in that incident, the board dismissed Altman on grounds that he was "not consistently candid in his communications with the board," but under the enormous pressure of employees collectively threatening to resign and investors like Microsoft, Altman was reinstated within days, and the board underwent major restructuring. The Superalignment team had originally been promised 20% of OpenAI's compute resources, but reports indicate this promise was never fully honored, with the team repeatedly hitting walls when competing for resources. Connected together, these events paint a picture of safety work being continuously marginalized within the organization.

The loss of this talent is not just OpenAI's loss — it may also mean that the company's internal safety checks and balances are being continuously weakened. When the people who best understand the risks choose to leave, the decision-making vacuum they leave behind may be filled by commercially driven judgment.

The Case Is Still Ongoing, But the Alarm Bells Are Already Ringing

The Musk v. Altman case is still being tried, and Murati's testimony is just one piece of evidence among many. Altman's side has not yet provided a detailed response to this specific accusation. But regardless of the final verdict, this trial has already exposed deep-seated problems in AI industry safety governance to public view.

At a time when AI capabilities are advancing at breakneck speed, who guards the safety baseline? When even internal safety gatekeepers say they cannot trust the top decision-maker, the urgency of external regulation and public oversight is plain for all to see. This trial may become a turning point in the history of AI industry governance — it forces the entire industry to confront an uncomfortable reality: in the race toward AGI, safety commitments without institutional safeguards risk becoming nothing more than empty words.

Key Takeaways

• OpenAI's former CTO Murati testified under oath in the Musk v. Altman case, accusing Altman of lying to her about AI model safety standards
• Altman allegedly falsely claimed the legal department had determined a new AI model did not require additional safety review, attempting to bypass multiple internal safety processes including red teaming and alignment evaluation
• Murati explicitly stated she cannot trust Altman's words, revealing a deep trust crisis between OpenAI's management and technical leadership
• The testimony provides powerful insider support for Musk's accusations that OpenAI has betrayed its nonprofit mission, with the case centering on the core dispute of OpenAI's transformation from a nonprofit to a for-profit company
• The incident reflects the fundamental tension in the AI industry between safety and speed, and the serious challenge of internal safety mechanism fragility in the absence of mandatory external regulation