Lilian Weng, co-founder of Thinking Machines Lab, has left the startup she helped build and rejoined OpenAI, according to TechCrunch. The move, reported July 29, 2026, follows Weng's public disclosure that the workload at the startup had taken a physical toll on her health.

Weng announced her departure in an internal Slack message at Thinking Machines and posted the same message on X. She wrote that the stress and workload at the startup had pushed her beyond what her health could sustain. Business Insider reported that Weng had tried to rest and take sick leave while Thinking Machines was racing to release Inkling, the company's first product.

The circumstances are unusual. A co-founder leaves a startup valued at $9 billion over health concerns, then quickly reappears at the company she had previously left. OpenAI confirmed to TechCrunch that Weng would be rejoining, and an OpenAI spokesperson said she will lead a top-level team focused on accelerating the company's internal research, with specific support for cross-research work on recursive self-improvement — the pursuit of AI systems that can iteratively improve their own capabilities.

Weng is no ordinary hire for OpenAI. She previously served as VP of AI Safety Research there and was listed as a member of OpenAI's Safety and Security Committee with the title "Head of Safety Systems" when the committee was formed in May 2024 (OpenAI). She stepped down from OpenAI in September 2024 (Bloomberg), then co-founded Thinking Machines Lab with Mira Murati, the former CTO of OpenAI, when the startup was founded in 2025 (Bloomberg). Thinking Machines has more than 30 people listed as members or employees and has attracted notable talent, including Andrew Tulloch among its listed members (Bloomberg).

Murati responded publicly to Weng's X post, writing: "I'm glad that you're putting your health first. Thank you for everything." TechCrunch reported that it is unclear whether Murati was aware Weng would rejoin OpenAI when she posted that reply.

That ambiguity is worth pulling on. Weng's stated reason for leaving was health, and she was direct about the cause being startup conditions. Whether OpenAI's environment will prove more sustainable is an open question, but the irony of leaving a startup for health reasons and joining a company known for its own intensity is hard to overlook. OpenAI's assignment, though, suggests the return is not a lateral move. Recursive self-improvement sits at the intersection of capability research and safety research, and it is one of the most consequential open problems in the field. Placing someone with Weng's safety credentials in a role designed to accelerate that work signals that OpenAI sees safety expertise and capability acceleration as complementary rather than oppositional.

For Thinking Machines, losing a co-founder during the push to ship its first product is a meaningful loss. The company launched in 2025 with a $9 billion valuation and a roster of senior researchers. Losing a co-founder to burnout during a product sprint, whatever the downstream personnel arrangements, is a data point about the human cost of the current AI startup cycle. Murati's public gracefulness in acknowledging the departure is notable, but the question of what she knew about Weng's next move, and when, remains unresolved.

The broader context here is about how the frontier AI talent pool operates. Weng's trajectory — from OpenAI safety leadership to startup co-founder and back to OpenAI in a role that blends safety and capability acceleration — captures something about the current moment. The talent pool at the frontier of AI research is small enough that the same names circulate across the most prominent labs, and the boundaries between safety and capability work are narrowing as those systems grow more capable. Where safety research once functioned largely as a check on capability work, the frontier now has safety expertise embedded directly into the acceleration of that work.