Andrew Tulloch, one of the AI industry’s most fought-over researchers, has resigned from Meta less than a year after joining the company in a deal reportedly worth as much as $1.5 billion. His next stop hasn’t been officially confirmed, but a X account @ArfurGrok, which has previously surfaced accurate AI personnel moves, claims Tulloch has already joined Anthropic. If confirmed, it would be a striking twist in an industry where Meta has spent the past year and a half aggressively poaching talent from Anthropic, OpenAI, and Google DeepMind — with Tulloch himself being one of Meta’s most expensive gets.
The Recruitment That Made Headlines Twice
Tulloch’s name first went viral in August 2025, when he turned down what The Wall Street Journal reported was a $1.5 billion, six-year compensation package personally dangled by Mark Zuckerberg. At the time, Tulloch was a co-founder and chief architect at Thinking Machines Lab, the AI startup launched by former OpenAI CTO Mira Murati. Zuckerberg had first tried to acquire Thinking Machines outright; when Murati turned that down, he shifted to picking off individual employees, and Tulloch was reportedly his top target. A Meta spokesperson dismissed the $1.5 billion figure as “inaccurate and ridiculous,” while maintaining that any large offer was tied to stock performance.
Two months later, in October 2025, Tulloch reversed course and joined Meta anyway — a move Thinking Machines confirmed to the WSJ, saying only that “Andrew has decided to pursue a different path for personal reasons.” The about-face, coming just weeks after Thinking Machines raised $2 billion at a $12 billion valuation and shipped its first product (Tinker, an API for fine-tuning AI models), stunned much of the AI industry. His return to Meta was seen as part of CEO Mark Zuckerberg’s broader campaign to staff Meta Superintelligence Labs (MSL) with elite research talent, alongside hires like Scale AI’s Alexandr Wang.
Less than a year later, he’s out again.
Career Trajectory
Andrew Tulloch is Australian, and studied mathematics at the University of Sydney, graduating with First Class Honours and the University Medal, before earning a master’s in mathematical statistics from the University of Cambridge. Before his AI career, he briefly worked in finance, interning at Goldman Sachs while still an undergraduate.
He joined Facebook (now Meta) in 2012 and spent roughly 11 years there, rising to become a key figure in the company’s large-scale machine learning systems and a core contributor to PyTorch, the open-source deep learning framework that underpins much of the modern AI industry — a reputation that earned him the nickname “the King of PyTorch” in Chinese tech media coverage of his moves. Notably, OpenAI co-founder Greg Brockman tried to recruit him as one of OpenAI’s first employees as early as 2016, but Tulloch turned the offer down at the time over a steep pay cut from his Facebook salary.
Tulloch left Meta in 2023 to join OpenAI, where he worked on ML systems, including pretraining for GPT-4o and GPT-4.5 and reasoning work on the o-series models. In February 2025, he left OpenAI to co-found Thinking Machines Lab alongside Mira Murati, taking on the role of chief architect. That stint lasted about eight months before he rejoined Meta in October 2025 — and now, roughly eleven months after that, he has left Meta again.
Part Of A Bigger Talent Shuffle
Tulloch’s exit lands amid a broader reshuffling of top AI talent across the industry’s biggest labs. Two other Thinking Machines co-founders, Barret Zoph and Luke Metz, have also recently left — Zoph moved to Google, while Metz joined Meta. Google’s chief scientist Jeff Dean and several senior researchers have also departed to start their own ventures.
Meta’s compensation-fuelled hiring spree — including nine-figure and, in Tulloch’s case, ten-figure packages — has been central to CEO Mark Zuckerberg’s push to catch up with OpenAI, Anthropic, and Google DeepMind in the race toward advanced AI. Tulloch’s resignation, coming just a day after Meta launched a new AI agent for consumer tasks like shopping and email, suggests that money alone isn’t guaranteeing loyalty in an industry where the pool of elite researchers remains vanishingly small.
We’ll update this piece as Meta, Anthropic, or Tulloch confirm further details.