Andrew Tulloch, the AI researcher whose recruitment to Meta drew attention for a compensation package reportedly worth as much as $1.5 billion over six years, is leaving the company after roughly one year to join Anthropic1,2,5.
Tulloch delayed his departure until Meta finished shipping Muse, the company's new family of open-source AI models and AI assistant announced this week4. He will start at Anthropic next week, joining its inference and performance team, which covers how models are trained and how they respond to users once deployed, according to Business Insider3.
Inside Meta's Superintelligence Labs
At Meta, Tulloch worked inside the TBD lab, a unit within the company's Superintelligence Labs division, reporting to Meta AI chief Alexandr Wang, the co-founder and former head of Scale AI6. CEO Mark Zuckerberg reportedly recruited Tulloch personally, offering a package that could have reached $1.5 billion over six years depending on bonuses and stock performance. Meta spokesperson Andy Stone denied that figure, calling it inaccurate and ridiculous.
Before returning to Meta, Tulloch spent 11 years at the company, left in 2023 for OpenAI, and then co-founded Thinking Machines Lab with former OpenAI CTO Mira Murati. Zuckerberg reportedly tried to acquire Thinking Machines Lab outright; when that failed, he approached its staff directly. Tulloch was among those who took the offer and rejoined Meta in 2025.
Timing and the Muse launch
The departure came days after Meta rolled out Muse, an agentic AI platform with coding and computer-use abilities pitched to everyday users rather than developers. Semafor, which broke the story on September 9, 2026, reported that Tulloch had been delaying his exit until the launch shipped, suggesting the move had been in motion for some time.
Semafor reported it is unclear exactly why Tulloch left. His personal website describes his Meta work in three words: "working on superintelligence".
ANALYSIS The departure underscores the difficulty of retaining elite AI researchers even at the compensation levels Meta has been willing to offer. Tulloch's move to Anthropic's inference and performance team rather than back to a startup or to another frontier lab places a researcher with deep systems-level experience at a direct competitor focused on model deployment efficiency.