# Lilian Weng becomes the fourth cofounder to leave Mira Murati's Thinking Machines Lab > Source: > Published: 2026-07-28 13:12:53+00:00 *The former OpenAI VP of AI Safety cited seven months of health problems worsened by startup stress, making her the fourth of six original cofounders to exit since Thinking Machines launched in February 2025.* Lilian Weng didn't dress it up. In a note shared with the team at Thinking Machines Lab, the former OpenAI VP of AI Safety said she'd been sick more in the past seven months than at any other point in her life, that startup-level pace had made things worse, and that she couldn't give 100% to her responsibilities. July 29 is her final day. She said she wants a more scoped role at a calmer company, though she remains committed to AI safety research. That candor is unusual. Most cofounder exits in AI come wrapped in language about "new chapters" and "exciting opportunities." Weng just said she's unwell and that the job made it harder to recover. Mira Murati, for her part, responded with grace, saying publicly that the Thinking Machines experience had been precious and that she was glad Weng was putting her health first. ## Four of six gone The numbers here are striking. Murati launched Thinking Machines in February 2025 with six cofounders and a vision for enterprise AI that adapts to individual businesses rather than imposing one general-purpose model on everyone. Within months, the attrition began. Andrew Tulloch left for Meta in October 2025. Barret Zoph and Luke Metz departed in January 2026, both returning to OpenAI, with Zoph's exit reportedly turning acrimonious after he was fired for sharing confidential company information with competitors. Murati subsequently named Soumith Chintala as the new CTO. Now Weng makes four. Two of the original six remain. The company isn't standing still despite the revolving door. Thinking Machines shipped its first open-source model, Inkling, on July 15, just two weeks before Weng's departure. Inkling is a 975 billion total parameter mixture-of-experts model, with 41 billion active parameters per query, trained on 45 trillion tokens spanning text, image, audio, and video. It carries an Apache 2.0 license. Companies can download, fine-tune, and commercialize it freely. The strategic pitch is pairing Inkling with Thinking Machines' customization platform, Tinker, and taking a revenue-sharing cut on enterprise deployments. It's a credible product play. But shipping a flagship model right as a cofounder exits for health reasons is a strange kind of milestone. The broader context makes Weng's departure more than a personnel story. Thinking Machines raised $2 billion at a $12 billion valuation, a seed round that made it one of the most richly funded AI startups before it had shipped anything publicly. That valuation created enormous expectations, and expectations at that altitude tend to produce exactly the kind of relentless pace Weng is describing. The competition for elite AI researchers is fierce: Meta, OpenAI, Google DeepMind, and others are all writing large checks to pull talent back or poach it. Tulloch went to Meta. Zoph and Metz went back to OpenAI. The gravity of those incumbent labs is real. According to reporting by Fortune and TechCrunch, the departures from Thinking Machines have involved money, compute constraints, and a lack of clarity on products and business model, not merely better offers elsewhere. That's a harder problem for Murati to solve than compensation alone. ## What Weng's exit actually means Weng spent nearly seven years at OpenAI, where she led the safety team and became one of the most cited researchers in the field. Her blog on neural networks and deep learning is a reference text - practitioners have used it for years. She does not leave positions lightly. That directness in her note, the admission that she's constantly worried about her health, is a real statement from someone who built an entire career around rigorous, sustained intellectual work. Founder burnout isn't a new story in tech, but it lands differently when it happens this visibly, this early, at a company operating at this valuation. The question Murati now faces isn't only how to replace a fourth cofounder. It's whether the operating culture that produced four exits in seventeen months can attract and hold the kind of talent Thinking Machines' ambitions require. Frankly, the product is compelling. The cofounder attrition rate is not. **Also read:** [Y Combinator's bet on killing tokenmaxxing just landed a $13.5M seed round as corporate America calls time on AI waste](https://startupfortune.com/y-combinators-bet-on-killing-tokenmaxxing-just-landed-a-135m-seed-round-as-corporate-america-calls-time-on-ai-waste/) • [Microsoft, Uber, and Commonwealth Bank confirm AI is cutting customer service jobs not just in theory](https://startupfortune.com/microsoft-uber-and-commonwealth-bank-confirm-ai-is-cutting-customer-service-jobs-not-just-in-theory/) • [Amazon just killed 20 AWS AI services it launched two years ago to chase enterprise deployment](https://startupfortune.com/amazon-just-killed-20-aws-ai-services-it-launched-two-years-ago-to-chase-enterprise-deployment/)