Why Lilian Weng Left Her Startup for Health and Returned to OpenAI

21CTO digest: Lilian Weng left the company she co‑founded, Thinking Machines, citing health, and quickly rejoined OpenAI, sparking buzz in the tech community.

Weng previously served as OpenAI’s Vice President of AI Safety, shaping the organization’s internal safety strategy. In 2025 she departed to co‑found Thinking Machines, a startup dedicated to frontier AI‑safety research, assembling top security researchers to challenge the safety paradigms of large models.

In an internal email she wrote, “I need to focus on my health first; it matters more than any code,” explaining that the high‑intensity workload had taken a toll on her well‑being, prompting her to step away for recovery.

OpenAI CEO Sam Altman announced her return as chief AI‑safety advisor, calling her “one of the most insightful voices in AI safety” and expressing enthusiasm about her helping steer the company’s safety research into a new phase.

The article notes that top AI leaders frequently move between large platforms and startups. Big firms offer resources, data, and influence, while startups provide greater freedom and potential high rewards, but also bring steep pressure—especially in AI safety, where responsibility is immense.

This personnel shift underscores the scarcity of experts who truly understand cutting‑edge model vulnerabilities and can devise robust defenses. Weng’s return is seen as a significant boost for OpenAI’s “super‑alignment” project.

Weng’s story illustrates the systemic pressure on pioneers in the high‑risk, fast‑paced AI race: as model capabilities expand, human physiological and emotional limits do not, suggesting the industry needs more sustainable work cultures so top talent need not choose between mission and health.

Open questions remain about how Thinking Machines will adjust after losing its core figure and what new role Weng will play at OpenAI, reflecting AI safety’s evolution from a support function to a core competitive advantage.