Google's AI shakeup: DeepMind's Hassabis steps aside, senior scientists depart
Ars Technica Ryan Whitwam ● Covered by 18 sources
Hassabis steps back from DeepMind, and four top Google AI researchers just quit. More brain drain at Google, just as its next big model runs late.
Based on reporting by Ars Technica, Ryan Whitwam — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Demis Hassabis, the man who's spent years telling anyone who'll listen that AGI is close at hand, is stepping away from the day-to-day grind at DeepMind. He's not leaving Google — he's shifting into an oversight role at Alphabet, becoming DeepMind's chair and the parent company's chief scientist, while still overseeing Isomorphic Labs, the AI-for-biomedicine effort he also runs. Running the actual research shop now falls to Koray Kavukcuoglu, who becomes DeepMind's senior vice president and reports straight to Sundar Pichai.
That reshuffle alone would be notable. But it landed at the same moment four other heavyweight researchers announced they're walking out the door entirely. Quoc Le, Oriol Vinyals, Jeff Dean, and Sanjay Ghemawat are leaving to launch a startup called Discovery Loop, aimed at using AI to automate scientific discovery. Dean isn't just any departure — he was Google's 30th employee, a guy whose fingerprints are on the systems that turned Google search into what it is, and someone employees half-jokingly mythologize with a running list of 'Jeff Dean Facts.' He and Vinyals also helped shape the Gemini models. According to Wired's Steven Levy, the new venture drew funding largely on the strength of its founders' reputations, and it's structured as a public benefit corporation — the same setup OpenAI and Anthropic use, for whatever that's worth.
Pichai reportedly fought to keep them, sitting through multiple meetings with Dean and Ghemawat before they confirmed they were leaving anyway. And this isn't an isolated blip. Earlier this summer, Noam Shazeer — a former Google VP of engineering, Gemini co-lead, and lead author of the 2017 'Attention Is All You Need' paper that introduced the transformer architecture underpinning modern generative AI — left for OpenAI. DeepMind VP John Jumper, known for building the AlphaFold protein-folding system, jumped to Anthropic around the same time, followed shortly by senior DeepMind researchers Jonas Adler and Alexander Pritzel, both described as key to getting Gemini off the ground.
Pichai's official line leans on Google's scale: strong talent, serious compute, and products already reaching more people than any rival. That's true as far as it goes. But it arrives as Google's momentum looks shakier than the PR framing suggests. At I/O this spring, the company promised Gemini 3.5 Pro would ship in June. Months later, all that's actually arrived are smaller, lower-power Flash models, while OpenAI and Anthropic keep shipping new systems of their own.
My take — AI-written commentary, not fact-checked reporting
Executives leaving isn't news; a wave of the people who actually built your flagship product walking out the same week your next big model is late is a pattern, not a coincidence. Google can talk up its compute and its talent bench all it wants, but reputation-funded startups and rival labs don't lure away a company's most respected engineers when things are going great inside. Watch where Dean, Jumper, and Shazeer end up shipping next — that'll tell you more about where the real AI momentum sits than any Pichai memo will.
Read more about this at: Ars Technica
