Thinking Machines co-founder Lilian Weng left for health reasons, then immediately rejoined OpenAI to lead recursive self-improvement research

Lilian Weng ran safety research at OpenAI. Now she is going back to lead a recursive self-improvement team.

On July 27, Weng shared an internal Slack message on X announcing her departure from Thinking Machines Lab, the startup she co-founded with former OpenAI CTO Mira Murati. She wrote that "the amount of consistent stress and workload" had pushed her beyond what her health could sustain 1. She was seeking, in her words, "a scoped role in a more predictable place, and not as a cofounder" 2.

Two days later, OpenAI confirmed she is rejoining the company to lead a new team. An OpenAI spokesperson told TechCrunch that the team will "support cross-research work on recursive self-improvement," a process that would allow an AI system to iterate on itself to become more powerful 1. Business Insider, also citing an OpenAI spokesperson, described the mandate as AI models training and improving their own successors 2.

What is new here is not the concept. It is the institutional commitment: a named team at a frontier lab, an explicit mandate to accelerate internal research, and a leader with a defined portfolio. TechCrunch reported in November 2024 that her title at OpenAI was VP of research and safety 3. Business Insider noted that she led the company's Safety Systems team and contributed to GPT-4's pretraining data, model safety, and evaluations 2.

The decision to place a safety researcher in charge of a team explicitly tasked with accelerating internal research carries two readings. In one, it is the most responsible choice available: the person who understands how AI systems fail is best positioned to oversee the systems designed to improve themselves. In the other, a safety pedigree functions as a credibility layer over a capability push, lending institutional trust to research whose stated goal is speed.

Neither reading can be confirmed from outside OpenAI. What is verifiable is that the team exists, the mandate is acceleration, and the leader has spent her career studying conditions under which AI systems behave unpredictably.

For Thinking Machines Lab, Weng's departure deepens a pattern. Of the startup's six co-founders, only two remain: Murati and chief scientist John Schulman. Barret Zoph and Luke Metz left for OpenAI; Andrew Tulloch left for Meta 2. Murati responded publicly to Weng's announcement, writing, "We'll miss you, it's been wonderful building Thinky together. I'm glad that you're putting your health first" 1. TechCrunch reports that it remains unclear whether Murati was aware Weng planned to rejoin OpenAI 1.

Thinking Machines raised $2 billion at a $12 billion valuation largely on the strength of the founding team Murati assembled, before the company had shipped a product 4. Four of those six co-founders have now departed.

The signal worth watching is not who went where. It is what OpenAI is building with the talent it pulls in. The company has now committed institutional resources to recursive self-improvement with a named leader and a defined mandate. Whether that effort turns out to be the most carefully governed capability push in the industry or the fastest one may depend on whether Weng's safety instincts survive contact with an acceleration timeline.