Google DeepMind CEO Demis Hassabis steps aside as four senior researchers leave to found Discovery Loop
Demis Hassabis steps down as Google DeepMind CEO to become chief scientist at Alphabet, while Jeff Dean and three other senior researchers leave to found Discovery Loop, an AI startup focused on automating scientific discovery.
Google DeepMind, the AI research arm that has shaped much of the modern AI landscape, just underwent its most significant leadership restructuring since its 2014 acquisition. Demis Hassabis is stepping down as CEO to become chairman of the unit and chief scientist at Alphabet, handing day-to-day operations to Koray Kavukcuoglu, now senior vice president of Google DeepMind. Hassabis’s new role places him at the corporate parent level, a move that signals how central AI has become to Alphabet’s identity rather than a retreat from it.
At the same moment, four of the company’s most prominent AI researchers are leaving to start something new. Jeff Dean, Sanjay Ghemawat, Oriol Vinyals, and Quoc Le are departing Google to found Discovery Loop, a public benefit corporation focused on automating scientific discovery. The group represents a staggering concentration of research talent — between them, they have authored foundational work on large-scale machine learning systems, Transformer architectures, and the software infrastructure that made modern AI possible. Alphabet is a founding investor in the new venture, a detail that suggests the corporate parent sees Discovery Loop less as a competitive loss and more as a strategic placement of talent into a mission too specific to contain within a general-purpose research division.
Kavukcuoglu’s promotion to daily operational leadership is significant in its own right. He has been a central figure in DeepMind’s technical direction for years, most recently leading its research into generalist AI agents. Handing him the reins while Hassabis shifts to a broader scientific mandate at the Alphabet level is the kind of move a corporation makes when it wants to keep a research organization’s technical culture intact while its founder takes on a wider brief.
The Discovery Loop departures, though, are the more structurally interesting part of the story. A public benefit corporation charter explicitly names a social mission alongside shareholder returns, and automating scientific discovery — not building better chatbots, not optimizing ad targeting — is about as far from the core Google business as a DeepMind-adjacent project can get while still being plausibly commercially relevant. The new company’s focus, as reported by Ars Technica, is precisely this: building AI systems that accelerate the pace of scientific research itself.
What’s less clear from the announcements is whether this represents a planned, orderly transition or something more abrupt. A founder stepping into a chairman role while a trusted lieutenant takes over operations is a classic succession pattern, and Alphabet’s founding investment in Discovery Loop — alongside the departure of four researchers whose combined institutional knowledge is difficult to replace — points toward an arrangement that was negotiated rather than reacted to. But leadership changes at this level, happening simultaneously, rarely reflect a single coherent story.
For Google, the practical question is what DeepMind’s research pipeline looks like without Dean, Ghemawat, Vinyals, and Le, all of whom have been central to its most significant technical achievements over the past decade. For Discovery Loop, the question is whether a public benefit corporation structure, even one seeded by Alphabet’s capital, can sustain the kind of long-term, high-risk research that automating science will require. Both questions will take years to answer.