Demis Hassabis quitte la direction de DeepMind pour se consacrer à l’AGI

The most prominent AI researcher in the world just reorganized his career around a single conviction: artificial general intelligence is near enough to demand his complete attention. Demis Hassabis, Nobel laureate and co-founder of DeepMind, stepped down as CEO of Google’s AI research lab to become its chairman and Alphabet’s chief scientist — a shift that removes him from daily operations and places him in front of what he calls the most important scientific question in history.

In a message to DeepMind staff, Hassabis said he has “been working towards AGI my whole life and now, like many of you, I feel it is close at hand.” Hearing this from someone who spent three decades building the actual systems — rather than commenting from the outside — changes how the claim lands. Most predictions about imminent AGI come from investors or commentators. Hassabis is the person who wrote the code.

Koray Kavukcuoglu, a 13-year DeepMind veteran who served as the lab’s chief technology officer, takes over as senior vice president, reporting directly to Alphabet CEO Sundar Pichai. The transition extends further than one role. Jeff Dean — a 27-year Google veteran and one of the most influential engineers in the history of machine learning — is departing to co-found Discovery Loop, a public benefit corporation focused on automating scientific research, alongside Sanjay Ghemawat, Oriol Vinyals and Quoc Le.

The move is striking because it frames AGI not as a distant target but as an imminent engineering challenge — one Hassabis believes requires the kind of focused personal attention that running a 4,000-person research organisation cannot provide. He will work directly with Pichai on what he described as “strategic and global AGI matters,” spanning every frontier model Alphabet builds. For anyone who has followed his career, the phrasing carries weight: Hassabis co-founded DeepMind in 2010 specifically to pursue AGI, and spent 15 years building toward it. He is not the kind of person who makes this statement lightly.

None of this makes the claim true. “Close” in AI research timelines has historically proven elastic, and DeepMind faces real operational friction: Gemini 3.5 Pro has been delayed months past its original June launch target, and the lab has lost several senior researchers to OpenAI and Anthropic over the past year. The reorganisation could equally be read as a consolidation triggered by competitive pressure as much as a signal of proximity to AGI. Alphabet shares fell roughly 5% on the news — markets clearly uncertain how to price a CEO who leaves because his goal is almost achieved.

Kavukcuoglu’s first task is Gemini 4, which Hassabis cited as evidence of “great progress” in his farewell message to staff. Discovery Loop, the new public benefit corporation founded by Dean and colleagues, expects to publish its first results on automated machine learning and scientific research before the end of the year. Hassabis, for his part, will now spend his days on the question he has been building toward since 2010.

Étiquettes: Google DeepMind, Demis Hassabis, IAG, intelligence artificielle générale, Jeff Dean