Mira Murati Warns AI Needs Constant Human Collaboration, Not Just Checkpoints
Benzinga reported Saturday that former OpenAI chief technology officer Mira Murati, now co-founder and CEO of Thinking Machines Lab, delivered a pointed warning about the trajectory of frontier AI at the Bloomberg Tech 2026 conference in San Francisco. Her core argument was direct. Human involvement in AI cannot be reduced to a simple sign-off at predefined stages.
The Tandem Bike Framework
Murati described the necessary relationship between humans and AI as a tandem bicycle. Both parties must contribute effort throughout the journey. She said the popular phrase “humans in the loop” gives a false sense of security. It implies a discrete checkpoint rather than ongoing, shared responsibility. “Both people are pedaling,” she told the audience, “but when you’re going up a hill, whoever is stronger pedals harder.” The point was that collaboration adjusts dynamically. It does not stop and restart.
Her newly launched company has built this philosophy into its earliest products. Thinking Machines Lab recently unveiled interaction models that process audio, text and video simultaneously in real time. Murati presented those systems as a practical expression of her tandem-bike vision, not just a rhetorical one.
Background: A High-Profile Departure From OpenAI
Murati left OpenAI in September 2024 after several years as its CTO, departing alongside other senior figures during a period of significant internal turbulence at the company. She subsequently founded Thinking Machines Lab in early 2025, raising substantial funding to pursue her own approach to AI development. Her departure came shortly after the boardroom crisis that briefly saw OpenAI CEO Sam Altman removed and reinstated within days.
Also Read: OpenAI Restructuring Plans Draw Scrutiny From State Regulators
Governance Over Individual Authority
Beyond the technical framing, Murati made a structural argument. She said AI’s future is not predetermined. Neither the catastrophic nor the idealistic outcomes many predict are inevitable. What matters, she argued, is building institutional checks rather than relying on the good intentions of any single leader or organization. Transparency and governance structures carry as much weight as the values of the people writing the code.
Her sharpest observation concerned timing. Pulling humans further out of the development process now, she warned, makes the alignment problem significantly harder to solve later. Getting the architecture of collaboration right at this stage matters more than many in the industry acknowledge.
Read Next: AI Governance Is Moving From Boardrooms to Capitol Hill
