Mira Murati, co-founder and CEO of Thinking Machines Lab, said frontier AI must function like a “tandem bike,” with both humans and machines collaborating throughout development.

Speaking at Bloomberg Tech 2026 in San Francisco on Friday, the former OpenAI CTO pushed back on the idea that “humans in the loop” is sufficient framing for safe AI development. “It sounds like a checkpoint where we’re signing off something and then you’re good to go,” she said.

“Both people are pedaling, but when you’re going up a hill, whoever is stronger is pedaling harder — but both hands are on the wheel,” Murati said. “That’s a system designed for collaboration.”

Not A Predestined Outcome

Murati said AI’s future is “not a predestined outcome,” dismissing predictions of either dystopia or utopia as oversimplified. She called for institutional checks and balances over individual authority, saying governance and transparency matter as much as the character of those building the technology.

Lessons From The Board Crisis

Murati, who served as interim CEO during OpenAI’s November 2023 board crisis, has seen the cost of weak institutional design firsthand. She said the conversation around AI leadership gets “too wrapped up” in individual character without enough focus on decision-making structures. “Even people that are well-intentioned can make mistakes,” she said.

The Albanian-American technology executive argued that pulling humans out of AI development now makes alignment harder to achieve later. “I see very little future possibilities that we can get this right when AI systems are even more capable,” she said.

Earlier in the interview, Murati pointed to Thinking Machines’ recently unveiled interaction models as an early demonstration of the company’s focus on human-AI collaboration. The models continuously process audio, text and video in real time, a philosophy she later described as a “tandem bike” in which humans and AI work together rather than operate independently.

Photo Courtesy: Shutterstock

Disclaimer: This content was partially produced with the help of AI tools and was reviewed and published by Benzinga editors.