According to an exclusive Semafor investigation, AI researcher Andrew Tulloch is leaving Meta just days after the launch of the Muse AI assistant and a new family of open-source models. The reason for his departure and his new plans have not been announced.
What happened
Semafor cites a person familiar with the situation, who says that Tulloch delayed his departure until the successful launch of Meta's new family of open-source models and the Muse assistant, announced on September 8. The publication outlines two scenarios for the future: a top position at one of the other frontier labs, OpenAI, Anthropic, or Google, or raising billions of dollars for his own startup on the model of Ilya Sutskever and Jeff Dean. Previously, media reported that Mark Zuckerberg personally invited Tulloch to Meta, offering a compensation package of around $1.5 billion over 6 years; The Wall Street Journal estimates the amount at over $1 billion, and the figures are disputed.
Context
Tulloch is a co-founder of the Thinking Machines Lab startup, a project by Mira Murati, and moved to Meta in 2025. There, he worked in the TBD lab, the superintelligence division under the leadership of Scale AI founder Alexandr Wang. None of the sources describe which specific models or areas he led. The story contains no technical statements: no new models, no benchmarks, no papers, no capability assessments — the novelty of the event is personnel-related. This is already the second move of a key researcher in a few years within the "revolving door" between frontier labs, where prices for top specialists reach billions of dollars.
Why this matters for the industry
For the industry, this is a signal about the talent market, not a product event: no new APIs, models, or benchmarks have appeared. Meta is losing one of the most expensive hires in the history of the AI industry from its superintelligence lab right after the Muse release, and the market reads this as two markers: the ceiling of how much Big Tech is willing to pay for a top specialist, and the vulnerability of the product line right after a major release. In each of the scenarios outlined by Semafor, a redistribution of forces occurs: a move to a frontier lab gives a competitor an already validated researcher, while his own startup will add a new well-funded player to the open models market. The observable indicator for the coming months will be the pace and quality of Meta's next open-source releases.
Why this matters for users
For the reader, the story is a cross-section of the "talent war": the co-founder of Thinking Machines Lab is leaving Meta right after the launch of Muse, meaning the company's flagship superintelligence lab has already lost a key figure. There are no direct practical changes: existing APIs, models, and prices are not changing, and there is nothing to put into production right now. What to watch for is the continuity of Meta's open-source model line and the Muse roadmap: if Tulloch assembles his own team, it will directly affect the open models race, and on a two-year horizon could produce a new frontier player with open models.
What is still unknown / limitations
Which specific models or areas Tulloch led in TBD, none of the sources describe. The amount of the compensation package is disputed: Semafor indicates around $1.5 billion over 6 years, The Wall Street Journal estimates it at over $1 billion. The story contains no technical claims — no papers, no architectures, no benchmark results — so any assessments of the impact on Muse and Meta's open-source family remain interpretation until official announcements.
Sources
- Semafor — Exclusive: AI researcher Andrew Tulloch is leaving Meta (Reed Albergotti)
- The Wall Street Journal — Star AI Researcher Andrew Tulloch Is Leaving Meta (behind paywall)
Author
Look at AI, editorial team