Lilian Weng steps down at Thinking Machines, then rejoins OpenAI days later

Lilian Weng’s tenure as co-founder of Thinking Machines ended almost as quickly as it began. She announced her departure this week, telling colleagues in a Slack message that the relentless pace of a startup had pushed her beyond what her health could sustain. “The amount of consistent stress and workload have pushed me beyond what my health can sustain physically,” she wrote.

Days later, the narrative took a sharp turn. OpenAI confirmed to TechCrunch that Weng is rejoining the company where she previously served as VP of AI Safety Research. Her new role puts her in charge of a top-level team accelerating internal research, with a specific focus on recursive self-improvement — the kind of AI that can iterate on its own capabilities. It’s a critical, deeply technical assignment at one of the most competitive labs on the planet.

The optics are a little jarring. Exiting one role due to health-sapping stress only to immediately resurface at another demanding research post invites questions. But the pressures of a co-founder are fundamentally different from those of a research lead, even one reporting to the top. You sleep differently when your name is on the lease, metaphorically speaking. Still, the speed of the boomerang underscores just how fiercely AI labs are fighting for proven talent.

Thinking Machines co-founder Mira Murati, formerly OpenAI’s CTO, responded to Weng’s initial departure post with warmth and support. “I’m glad that you’re putting your health first,” Murati wrote. It’s unclear whether Murati knew Weng was already walking back toward Altman’s orbit. The whole sequence feels less like a standard job hop and more like a stress fracture in the startup’s early foundation.