Anthropic Poaches Two Senior Google Researchers Tied to Gemini

Jonas Adler and Alexander Pritzel, two senior Google AI researchers, are planning to leave for Anthropic. Both were described as key contributors to Gemini, Google’s flagship AI model, and as highly regarded by their peers.

Adler’s work was tied to AI coding. Pritzel was involved in pre-training, the stage where models are trained on large quantities of data. Colleagues were “sad to see them go,” according to Julia Love, and the exits matter less as a numerical staffing loss than as a signal about who is leaving.

The two planned moves follow other high-profile departures that had already sharpened concern around Google’s AI ranks: Noam Shazeer moving to OpenAI, and John Jumper, a Nobel laureate, heading to Anthropic. Together with Adler and Pritzel, those exits raised questions about “who else may be on the way out.”

Ed Ludlow framed the central uncertainty as whether this amounts to a “mass wave of departures” that puts Gemini in trouble, or whether that overstates the case. Love drew a narrower boundary around what Bloomberg had reported: beyond Jumper and Shazeer, the new names were Adler and Pritzel. “We’re talking about a small number of people,” she said.

But that caveat does not make the losses immaterial. In frontier AI, Love said, the group of people who can “really move the needle on these models” is relatively small.

Google’s position, as relayed in the discussion, is that its bench of talent remains “very deep.” The tension is therefore not between collapse and irrelevance. It is between Google’s institutional depth and the fact that a small number of unusually capable researchers can have outsize importance to model progress.

Alphabet shares were already under pressure as the staffing news circulated. Ludlow noted that pressure on Google’s parent company, and the on-screen intraday panel showed Alphabet A-shares moving lower.

The panel read: “INTRADAY Alphabet A-Shares 341.58 3.71 1.07%.” The visual established the market backdrop for the segment, but it did not by itself demonstrate that the reported departures caused the share move.

The staffing pattern, rather than the stock chart, is the substantive issue. Anthropic is the principal beneficiary named in the reported moves: Adler and Pritzel are planning to go there, and Jumper was also described as heading there. Shazeer, by contrast, was cited as moving to OpenAI.

The account also included internal context around Noam Shazeer’s departure. The issue was compute, described as a precious and politically charged resource inside Google.

Google has substantial computing power, but it also has many competing demands on it: cloud clients, researchers, and widely used products. Shortly before Shazeer announced his departure, compute assigned to one of his projects was shifted to a larger pre-training team in London. The move “seems” to have been an attempt to increase collaboration, Love said, while emphasizing that Bloomberg was still piecing together exactly what happened.

That detail broadens the personnel story without turning it into a single-cause explanation. The source did not establish that compute allocation caused Shazeer’s exit, or that the same dynamic explains Adler and Pritzel’s planned departures. It did describe an operating environment in which even Google’s large infrastructure base is not effectively unlimited, because research projects, cloud customers, and major products all compete for the same critical resource.

For frontier-model teams, that competition matters. Talent and compute are linked inputs: the researchers who can materially improve models need access to the infrastructure required to train and test them. The reported departures therefore sit inside a broader constraint, where scarce people and scarce compute both shape the pace and direction of AI work.