Between 18 and 25 June 2026, five senior researchers departed Google DeepMind for rival labs. Four joined Anthropic; one joined OpenAI. The departures span the Gemini pre-training team, the Nobel-winning AlphaFold group, AI safety, and coding — and they have already moved markets, with Alphabet losing roughly $270 billion in market capitalisation across two trading sessions.
This is not routine attrition. It is a structural shift in where the most capable AI researchers choose to work, and what that means for the models shipping over the next 12 months.
TL;DR
- Noam Shazeer (Transformer co-author, Gemini co-lead) left for OpenAI on 18 June — less than two years after Google paid $2.7 billion to acqui-hire him back from Character.AI.
- John Jumper (2024 Nobel Prize in Chemistry, AlphaFold co-creator) announced his move to Anthropic on 19 June.
- Jonas Adler and Alexander Pritzel (Gemini contributors, former AlphaFold collaborators) are leaving for Anthropic, reported 24 June.
- Arthur Conmy (Gemini 2.5 post-training, AI safety) announced he is joining Anthropic on 25 June to work on alignment for upcoming Claude models.
- Alphabet stock fell approximately 7% on 22 June — its worst single session in over a year.
Who Left and What Did They Work On?
Noam Shazeer co-authored the 2017 paper "Attention Is All You Need," which introduced the Transformer architecture underpinning every major large language model today. He co-led Gemini development after Google acqui-hired him back from Character.AI in 2024 for $2.7 billion. He lasted under two years before joining OpenAI, where Sam Altman publicly welcomed him. For context on what OpenAI is building with this calibre of hire, see our OpenAI Jalapeno chip deep dive.
John Jumper spent nine years at DeepMind and won the 2024 Nobel Prize in Chemistry alongside Demis Hassabis for AlphaFold, which predicts over 200 million protein structures. He announced his departure to Anthropic on 19 June. Because Jumper is UK-based, lengthy non-compete agreements under British law mean he likely will not begin work at Anthropic until 2027.
Jonas Adler and Alexander Pritzel both contributed to Gemini and previously worked with Jumper on AlphaFold research. Bloomberg reported their departures to Anthropic on 24 June. Adler specialised in AI coding; Pritzel focused on pre-training and model training.
Arthur Conmy, a senior research engineer who worked on Gemini 2.5 post-training and AI safety/interpretability from 2023 to 2026, announced on 25 June that he is joining Anthropic to work on aligning upcoming Claude models during training. He previously worked at Redwood Research. For more on what Anthropic is building, see our coverage of Claude Opus 4.8 dynamic workflows.
Why Are They Leaving?
Three factors converge:
Equity upside. Anthropic recently raised funding at a $965 billion valuation and is considering going public as soon as autumn 2026. OpenAI confidentially filed an S-1 on 8 June targeting an $830 billion to $1 trillion valuation. Pre-IPO equity at these labs offers a financial incentive that Google's mature stock cannot match. Senior researchers joining now stand to gain substantially if either company lists successfully.
Autonomy and resource allocation. Bloomberg reported that computing power dedicated to one of Shazeer's projects was reassigned to a London-based DeepMind team shortly before his departure, part of an effort to streamline pre-training work and boost cross-team collaboration. For researchers accustomed to directing large compute budgets, that kind of reorganisation can feel like a demotion even if it is not intended as one.
Mission alignment. Conmy's move is explicitly about safety research — he stated he will work on aligning Claude models during training. Anthropic's positioning as the "safety-first" lab attracts researchers who want their interpretability and alignment work to directly shape products rather than sit in a parallel research track.
What Does the Google DeepMind Talent Exodus Mean for Gemini?
The immediate signal is the Gemini 3.5 Pro delay. As of 26 June, the model has been pushed to July for final adjustments. Google is reportedly expanding its AI coding strike team to cover "midtraining" — a response to the gap left by departing researchers.
A 2025 SignalFire analysis found that DeepMind engineers are 11 times more likely to leave for Anthropic than the reverse. That ratio suggests this is not a one-off event but a sustained directional flow.
Demis Hassabis responded publicly: "There's a lot of talent movement between all the leading labs and we win our fair share of the top talent. We have by far the biggest and broadest research bench of any of the labs out there." The statement is factually defensible — DeepMind still employs thousands of researchers — but breadth is different from depth at the frontier.
What Does This Signal for the Broader AI Industry?
The AlphaFold team that won the 2024 Nobel Prize is now largely scattered across competing organisations. Anthropic appears to be systematically targeting Google's Gemini team across science, coding, training, and safety verticals. This is a coordinated talent acquisition strategy, not opportunistic poaching.
For investors, the $270 billion market-cap wipe across two sessions shows the market takes researcher departures seriously — perhaps more seriously than product announcements. Alphabet fell roughly 7% on 22 June alone, its worst single-day performance in over a year.
The broader pattern: pre-IPO AI labs with equity upside are pulling senior talent away from Big Tech incumbents. Google, Meta, and Microsoft all face the same structural disadvantage — their stock is already priced for AI success, while Anthropic and OpenAI offer the prospect of 5-10x returns on equity grants. For more on OpenAI's trajectory, see our GPT-5/6 gated launch coverage, and for Anthropic's latest models, our Claude Fable 5 analysis.
FAQ
Q: How many researchers has Google DeepMind lost to competitors in June 2026? A: Five confirmed senior departures between 18 and 25 June: Noam Shazeer to OpenAI, and John Jumper, Jonas Adler, Alexander Pritzel, and Arthur Conmy to Anthropic.
Q: Will John Jumper start at Anthropic immediately? A: Likely not. UK-based researchers face lengthy non-compete agreements under British law, meaning Jumper probably will not begin work at Anthropic until 2027.
Q: Has the talent exodus affected Google's product roadmap? A: Yes. Gemini 3.5 Pro has been delayed to July for final adjustments, and Google is expanding its coding strike team to cover midtraining work left by departing staff.
Q: How much did Alphabet's stock drop following the departures? A: Alphabet fell approximately 7% on 22 June — its worst single session in over a year — with an estimated $250-270 billion in market capitalisation erased across two trading sessions.
Q: What is driving researchers to leave Google for smaller labs? A: Pre-IPO equity at Anthropic ($965 billion valuation) and OpenAI ($830 billion-$1 trillion target), greater autonomy over compute allocation, and closer alignment between research interests and product direction.
This article was produced with AI research assistance. See our editorial process at /how-we-work.
Discussion
0 comments