The drain of Google AI talent to rival labs picked up speed again this week, as two more senior researchers behind the Gemini model headed for the door. According to Bloomberg reporting summarized by TechCrunch, Jonas Adler and Alexander Pritzel are leaving Google for Anthropic. They follow a run of high-profile exits in June 2026 that has rattled investors and reopened a question the company would rather not face: whether the lab that helped invent the modern era of AI can hold onto the people who built it.
- Jonas Adler and Alexander Pritzel, both key contributors to Gemini, are the latest Google researchers to join Anthropic.
- They follow Nobel laureate John Jumper, who left Google DeepMind for Anthropic, and Gemini co-lead Noam Shazeer, who departed for OpenAI, both within the same stretch of June 2026.
- A SignalFire report found DeepMind engineers were nearly 11 times more likely to leave for Anthropic than the reverse, and Alphabet shares fell around 5 to 6 percent as the news landed.
The pattern is what makes this more than routine churn. Individual stars move between labs all the time. A cluster of departures concentrated at one company, hitting both its flagship language model team and its science division inside a few weeks, reads as something structural. The people leaving are not junior engineers; they are among the most decorated and load-bearing researchers in the field.
Who is leaving Google AI talent ranks
Start with the two newest names. Jonas Adler worked on AI-powered coding, one of the most commercially contested capabilities in the current model race. Alexander Pritzel focused on training AI systems, the core craft of turning raw compute and data into a working model. Both were viewed internally as important figures behind Gemini, Google's main answer to GPT and Claude. Losing two people central to that effort, to the same competitor, in a single announcement is not a coincidence anyone at Google will read comfortably.
They are the second wave. Days earlier, John Jumper left Google DeepMind for Anthropic after nine years at the lab. Jumper is not an ordinary hire. He shared the 2024 Nobel Prize in Chemistry with DeepMind chief executive Demis Hassabis for AlphaFold, the system that predicts the three-dimensional structure of proteins and reshaped computational biology. As The Decoder noted, Jumper's focus on protein structure lines up neatly with Anthropic's growing interest in AI for science, which gives the move a strategic logic beyond a paycheck.
The third departure may sting the most for symbolic reasons. Noam Shazeer, a co-lead of the Gemini models, is heading to OpenAI. Shazeer co-authored Attention Is All You Need, the 2017 paper that introduced the Transformer architecture underpinning nearly every large language model in use today. He had left Google once before to co-found the chatbot startup Character.AI with Daniel de Freitas, then returned in 2024 through a deal in which Google licensed Character's technology for a reported 2.7 billion dollars, a transaction widely read as a way to bring Shazeer back into the fold for Gemini. Two years later he is leaving again, this time for the company many see as Google's fiercest rival.
The Shazeer boomerang and what it cost
Shazeer's path through Google is worth tracing, because it shows how expensive retaining a single researcher can become and how fragile that retention can prove. He spent much of his career at Google, where he helped build LaMDA, the conversational system that prefigured the chatbot wave. He then left to co-found Character.AI, which grew into a viral consumer hit built on his own models. That success made him a target for re-acquisition.
In 2024 Google brought him back through an unusual arrangement. Rather than a conventional acquisition, the company licensed Character.AI's technology in a deal reported at 2.7 billion dollars, a structure that let Google secure the technology and the founders without buying the startup outright. The widely held reading was that the payment was, in large part, about getting Shazeer and his expertise back onto the Gemini effort. For a company to spend at that scale to reacquire one researcher's talent, then watch him leave for OpenAI two years later, is a vivid illustration of how mobile the very top of this field has become. Money bought time, not permanence.
The episode also reframes the current departures. If even a multibillion-dollar arrangement could only hold a researcher for two years, the structural pull of a rival on the cusp of an IPO is formidable. Retention at the frontier is not a problem a company solves once; it is a recurring negotiation against rivals whose offers keep escalating.
Why losing a Nobel laureate matters
Jumper's move deserves its own weight, because AlphaFold is arguably the most consequential applied AI achievement of the past decade. The system predicts how proteins fold into three-dimensional shapes from their amino-acid sequences, a problem that had resisted biologists for fifty years. Its predictions accelerated research across drug discovery, disease understanding, plus basic biology, and the released database of predicted structures became a standard tool in laboratories worldwide. The 2024 Nobel Prize in Chemistry recognized exactly that impact.
That is why his departure carries a meaning beyond one headcount. Jumper represents a proven ability to take AI out of the chatbot box and into a domain where it produces durable scientific value. Anthropic has signaled a growing focus on AI for science, and acquiring the person most identified with that frontier is a way to buy credibility and direction at once. For Google DeepMind, which has long positioned scientific discovery as central to its identity and its public justification, losing the face of its biggest science win to a direct competitor cuts at something the lab considers core to who it is.
The practical question is whether the AlphaFold lineage stays strong at DeepMind without him. Large scientific programs outlast any single leader, and the AlphaFold team is more than one person. But leadership and taste are hard to replace, and the researchers who define a breakthrough often carry forward an intuition that does not transfer through documentation alone.
Can DeepMind stay at the forefront
The departures have prompted a sharper version of the question, voiced in coverage such as a Fortune analysis asking whether Google DeepMind can remain at the forefront of AI development as its top talent walks. It is a fair question to raise, though an early one to answer. DeepMind continues to produce frontier models, and the Gemini line remains genuinely competitive on capability and price, which is not the profile of a lab in freefall.
What the question really probes is momentum. AI research has a strong herd quality. Talent flows toward whichever lab is perceived to be winning, and that perception can become self-fulfilling, drawing in the next generation of hires and starving rivals of the same pool. Google's challenge is to keep the perception of momentum on its side even as a visible slice of its roster relocates, because the story investors and recruits tell themselves about who is ahead has real consequences for who actually ends up ahead.
For now the honest assessment is mixed. Google retains scale, vast compute, plus a deep research organization, and a handful of exits does not erase any of that. At the same time, the specific people leaving are unusually significant, the direction of the flow is one-sided, and the financial pull of pre-IPO rivals is not going away soon. The next two quarters of hiring and shipping will say far more than this week's headlines about whether the lab holds its place.
The data behind the talent war
One striking departure can be dismissed as a personal decision. A directional flow shows up in the numbers, and the numbers favor Anthropic. SignalFire's 2025 State of Talent Report found that engineers at DeepMind were nearly 11 times more likely to leave for Anthropic than to move in the opposite direction. That is not parity with some noise around it; it is a lopsided current running one way.
The same body of data offers a clue about why the flow is so one-sided. Anthropic leads frontier AI labs in retention, reportedly keeping around 80 percent of its technical staff over a two-year window. A company that both attracts rivals' best people and holds onto its own builds a compounding advantage, because every senior hire who stays becomes a magnet for the next. Talent density is self-reinforcing, and right now the density is accruing at Anthropic rather than at the lab that trained many of these researchers.
For Google, the timing is unkind. The departures cluster exactly as Anthropic and OpenAI move toward public offerings. Pre-IPO equity is the single most powerful recruiting lever in technology, because it offers the prospect of a liquidity event that can dwarf even a large salary. Researchers who join a soon-to-be-public lab are betting that their equity will be worth far more after the offering than the compensation a mature company like Alphabet can match. When two rivals are both approaching that moment at once, the pull on a single incumbent is doubled.
Compensation is the stated driver, but it is rarely the whole story. Researchers also weigh autonomy, the speed at which their work ships, plus whether they believe their employer is on the winning path. Equity makes the decision easier to justify, but the sense of momentum is often what makes it feel right.
Why investors flinched
Markets noticed before most observers did. Alphabet shares fell roughly 5 to 6 percent on June 22, 2026, with market commentary tying the drop to twin concerns: the scale of Google's AI spending and its ability to retain senior AI talent. A few percent of Alphabet's market value is an enormous absolute number, and the fact that researcher departures could move it at all says something about how the market now prices human capital in AI.
That reaction reflects a shift in how investors think about these companies. For most of the software era, talent was important but fungible at the margin; a strong company could lose a few engineers and replace them. In frontier AI, a small number of people hold an outsized share of the know-how required to push a model to the frontier. When several of them leave for a competitor in a compressed window, the market reads it as a transfer of capability, not just headcount. The share-price move is the financial expression of that fear.
It is worth keeping the reaction in proportion. A 5 to 6 percent dip is a warning shot, not a collapse, and Alphabet remains one of the most resource-rich companies on earth with a vast compute fleet and a deep bench beyond the departed names. The signal is about trajectory and perception, not an immediate hit to Google's products. Whether it becomes a lasting problem depends on whether the outflow continues or proves to be a concentrated burst tied to the IPO window.
Google's response and its deeper bench
Hassabis pushed back on the narrative directly, asserting that Google still holds the deepest research bench of any AI lab. The claim is defensible. Google DeepMind employs thousands of researchers, controls one of the largest compute infrastructures in the world, and continues to ship competitive models, with Gemini 3 versions trading benchmark wins against the latest from OpenAI and Anthropic. No single set of exits empties a bench that deep.
The counterpoint is that depth is not the same as the frontier. Frontier progress tends to be driven by small, exceptional teams, and the loss of the specific people who led a breakthrough can slow the next one even when the broader organization remains large. AlphaFold was not built by a thousand people; it was built by a tight group around Jumper. The Transformer paper had eight authors, and Shazeer was one of them. The worry is not that Google runs out of researchers but that it loses the particular individuals who tend to be present when the big leaps happen.
Google also did not respond to press requests for comment on the latest departures, which leaves Hassabis's bench remark as the company's main public posture. Silence in the face of a developing story can read as confidence or as a lack of a ready answer, and which interpretation sticks usually depends on what happens next. If the outflow slows, the bench argument holds. If it continues into the second half of 2026, the absence of a fuller response will look like a missed chance to reassure both staff and investors.
What the moves say about Anthropic and OpenAI
The flip side of Google's loss is a deliberate accumulation by its rivals. Anthropic in particular has become the destination of choice, pulling in Jumper, then Adler and Pritzel, in quick succession. Jumper's arrival strengthens a stated push into AI for scientific discovery, a domain where Anthropic sees both societal value and a defensible niche away from the crowded chatbot market. Hiring the person most associated with AI's biggest scientific success to date is a clear statement of intent.
OpenAI's capture of Shazeer carries its own weight. Bringing aboard a co-inventor of the Transformer, and a co-lead of a direct competitor's flagship model, is both a capability gain and a morale signal in the talent market. Every prominent hire makes the next one easier, because researchers want to work alongside the people whose names they have read on the papers that defined the field. The labs approaching their IPOs are using that dynamic, and their war chests, to assemble concentrations of talent that compound over time.
None of this guarantees outcomes. A researcher who shone inside one organization's culture and tooling does not automatically reproduce that work elsewhere, and integration takes time. But the direction of travel is unambiguous, and for now it runs away from Google. The company that did more than any other to seed the current generation of AI is watching a meaningful share of that seed corn replant itself on competitors' ground.
The bigger picture for the AI race
Step back and this looks like a defining feature of the 2026 AI landscape rather than a one-company story. The frontier is being contested not only with compute and data but with a small, mobile population of elite researchers whose choices can shift the balance between labs. When that population concentrates, capability concentrates with it, and the gap between leaders and the rest can widen faster than spending alone would predict.
For Google, the path back to stability runs through retention as much as recruitment. Matching pre-IPO equity is hard for a public company, so the lab will have to compete on the things money cannot fully buy: the chance to do landmark work, ship it quickly, plus the feeling of being part of a winning effort. Those are exactly the levers Hassabis was reaching for with the deepest-bench claim. Whether they are enough to stem the flow of Google AI talent is the question the rest of 2026 will answer, and the early evidence suggests the company has work to do.