Mira Murati Returns to the Spotlight—With a New AI Startup and a Warning About Power

After 18 months largely out of view, Mira Murati sat down with Bloomberg’s Emily Chang in San Francisco on June 5. The former OpenAI chief technology officer, now CEO of Thinking Machines Lab, used the moment to preview technology built for real-time conversation. She also reflected on the turmoil that briefly made her interim boss at her old company—and issued a pointed caution about how few people control the direction of artificial intelligence.

Murati left OpenAI in September 2024. In the months since, her new startup raised $2 billion, secured access to Nvidia’s next-generation Vera Rubin systems, shipped one product, and watched several key researchers walk out the door. The industry she re-entered is fiercer, richer, and more concentrated than the one she left.

Thinking Machines Lab launched in February 2025 with a founding team drawn from OpenAI, Meta, and other labs—including John Schulman, Barret Zoph, Lilian Weng, Andrew Tulloch, and Luke Metz. Early funding valued the startup at $12 billion, led by Andreessen Horowitz. Investors bet on Murati’s track record: she played a central role in releasing ChatGPT, DALL-E, and Codex.

But progress has been slower than the valuation suggested. The company’s first offering, Tinker, arrived in October 2025—an API for fine-tuning open-source models. Forbes called it useful but not a breakout hit. In January 2026, Zoph, Metz, and Sam Schoenholz returned to OpenAI. Five other early employees joined Meta, lured by nine-figure compensation packages. Fortune covered the exits.

Murati addressed the departures directly. “When I wake up in the morning, I am not thinking about how to kill the competitor,” she told Chang. The line drew laughs but underscored a tension: OpenAI dominates headlines, Anthropic has raised tens of billions, and Elon Musk’s xAI keeps expanding.

The new technology she highlighted aims to change how humans and machines interact. Thinking Machines calls them “interaction models”—systems that handle continuous streams of audio, video, and text, updating every 200 milliseconds. The first example, TML-Interaction-Small, returns responses in 0.40 seconds. It supports full duplex interaction: users can interrupt, and the system adjusts midstream, picking up on pauses, corrections, and tone. TechCrunch detailed the approach.

Murati framed the work as a first step, positioning it beside Tinker as part of a broader effort to make AI feel more like a collaborator and less like a question-and-answer box. Her thesis: powerful systems require closer human partnership, not detachment. “Neither dystopian nor utopian futures are locked in,” she said. “The choices made right now will shape which path prevails.”

Chang asked about the episode still known inside OpenAI as “the blip.” In November 2023, the board fired Sam Altman, and Murati became interim CEO for five turbulent days. She stood by her actions, saying she felt clear about protecting the mission and the team. Without her involvement, she added, the company would have imploded. But she admitted room for improvement: she would have pressed for more information, a smoother transition, and greater transparency.

When asked whether she still trusts Altman, Murati sidestepped. Her worry centers less on any single personality and more on the concentration of consequential decisions in too few hands. “This pattern exists across the industry,” she said. “Good people still make bad calls. Well-intentioned organizations can drift. Structural checks remain scarce.” The critique lands at a moment when a handful of labs control the most advanced models, the largest compute allocations, and the biggest capital inflows.

Murati’s return arrives as the sector wrestles with speed versus caution. Her call for better governance doesn’t propose slowing progress. It argues for mechanisms that prevent drift. Her product direction tries to embed collaboration directly into architecture. The bet is that systems designed this way will prove more reliable partners.

Whether that vision can hold against competitors racing to scale remains unproven. Thinking Machines has one commercial offering, and its research team has thinned at the top. Yet Murati sounded measured. Heads-down execution still matters, she suggested. But silence carries its own risks. The industry has moved fast without her. Now she aims to help set its next direction—with real-time models in testing and a governance critique on record.