Lilian Weng Left Thinking Machines for Her Health. OpenAI Hired Her Back Within 48 Hours.
Lilian Weng co-founded Thinking Machines Lab with Mira Murati in February 2025. On July 29, 2026, she announced her last day, citing seven months of health problems the startup pace made worse. By the next morning, OpenAI confirmed she was returning to lead a new internal research team focused on recursive self-improvement.
The turnaround took less than 48 hours.
What Weng Said In Her Own Words
She didn’t soften it. Most AI departures come wrapped in vague language about new directions and personal growth. Weng posted the actual reason on X.
“I don’t feel I’m able to continue at the pace a startup requires,” she wrote in an internal Slack message she made public. “After thinking about it for several months, I ultimately have to admit that the amount of consistent stress and workload have pushed me beyond what my health can sustain physically.”
She said she had been sick more often in the past seven months than at any other point in her life. She didn’t name a specific diagnosis.
Mira Murati responded publicly. “We’ll miss you, it’s been wonderful building Thinky together. I’m glad that you’re putting your health first,” Murati wrote on X. Whether Murati knew about the OpenAI return at that point is unclear.
Who Is Lilian Weng?
She joined OpenAI in 2017. Her first project was Dactyl, the robotic hand that taught itself to solve a Rubik’s Cube. Over seven years, she moved from robotics research to building the company’s entire safety infrastructure.
By the time she left in 2024, she was VP of Research, Safety and had unified OpenAI’s safety efforts into a single team of over 80 scientists and engineers. She built the fine-tuning API, the moderation endpoints, and the applied safety frameworks that OpenAI’s products still run on today.
Business Insider put her on its 2024 AI Power List. That’s the track record she walked out with when she joined Thinking Machines.
What She’ll Do at OpenAI Now
OpenAI confirmed the hire to TechCrunch on Wednesday. A company spokesperson said Weng will lead a top-level team focused on accelerating internal AI research. The specific mandate is recursive self-improvement — building AI systems that can iterate on themselves to become more capable over time.
OpenAI’s own June 2026 governance blueprint described recursive self-improvement as “potentially the most consequential frontier safety issue of the coming decade.” They’re putting one of their best safety architects in charge of it.
That’s not a coincidence. It’s a deliberate structural choice.
At the startup, Weng carried everything a co-founder carries. At OpenAI, she gets a focused research role with clearer scope. Same cutting-edge work, different workload shape. If you want to understand how these AI systems actually learn to write better prompts and outputs, our guide to prompt engineering explains the underlying mechanics.
Thinking Machines Is Now Down to Two Original Co-Founders
This is the bigger story that the health angle almost overshadows.
Thinking Machines launched in February 2025 with six co-founders. Weng is the fourth to leave. Here’s the full exit list:
- Andrew Tulloch — left for Meta in October 2025
- Barret Zoph (CTO) and Luke Metz — left in January 2026, both returned to OpenAI. Murati publicly said Zoph was fired for misconduct. OpenAI disputed that account.
- Lilian Weng — left July 29, 2026, citing health
Only Murati and Chief Scientist John Schulman remain from the original founding group.
By May 2026, 13 of the company’s 42 founding members had left. That’s nearly one in three. Thinking Machines still holds a $12 billion valuation and a multi-year NVIDIA partnership that includes deploying at least one gigawatt of Vera Rubin systems starting in 2027. The resources are real. But the founding team that gave the company its early credibility is mostly gone.
A Pattern Nobody Is Talking About Loudly Enough
Weng’s exit is not isolated. It fits a visible pattern in the AI industry right now.
Fidji Simo stepped back from a senior OpenAI role this month after a flare-up of a chronic illness. In January 2026, xAI co-founder Greg Yang moved to an advisory role following a Lyme disease diagnosis. His public statement was blunt: he needed to “go founder mode” on his health.
The people stepping away tend to sit one level below the CEO. The bosses at the top mostly stay put. And the race doesn’t slow down for anyone.
Some teams inside top AI labs reportedly run 996 schedules 9am to 9pm, six days a week. Nobody publishes that number officially. But the pattern of exits tells you something real about what those schedules cost.
The AI tools that researchers like Weng helped build ChatGPT among them — are now used by hundreds of millions of people. The humans who built them are running out of steam. That tension isn’t going away.
What This Means Going Forward
Weng’s return strengthens OpenAI’s research depth at exactly the moment they need it most. Recursive self-improvement is not a niche research topic. If it works the way OpenAI believes it can, it changes how AI systems are developed at a fundamental level.
And for Thinking Machines, the question is simple: can Murati rebuild the founding team’s momentum with only one original co-founder beside her? The company has money and computing commitments locked in. What it’s losing is institutional knowledge and the research credibility that made its early hiring possible.
Weng walked out the door. She walked back into OpenAI within two days. That tells you something about where the real gravity still sits in this industry.
