Google's AI Brain Drain Deepens: Gemini Architects Adler and Pritzel Join Exodus to Anthropic

Google has built some of the most consequential AI research organizations in history. It invented the Transformer. AlphaFold changed our ability to understand life itself. DeepMind has produced more landmark Nature papers than any private company. And yet, in the span of a single month in June 2026, the company has watched an extraordinary cascade of its most important AI researchers walk out the door.

The latest departures: Jonas Adler and Alexander Pritzel, two senior researchers who played significant roles in developing Google’s Gemini models, are joining Anthropic. TechCrunch confirmed the moves on June 24 — adding two more names to what has now become an undeniable, historically unusual pattern.

The Exodus, Catalogued

The parade of departures from Google’s AI research organizations to its rivals has been striking in both pace and caliber:

Noam Shazeer is the most historically significant loss. A Google research fixture for 24 of the past 25 years, Shazeer is one of the co-authors of the landmark 2017 paper “Attention Is All You Need,” which introduced the Transformer architecture that underlies virtually every modern LLM. He left to co-found Character.AI; Google spent $2.7 billion to acqui-hire the company and bring him back in 2024. He has now departed again — this time for OpenAI.

John Jumper, Director at Google DeepMind, co-received the 2024 Nobel Prize in Chemistry for AlphaFold, the protein structure prediction system that has become one of the most impactful computational biology tools ever built. He’s heading to Anthropic. Losing a sitting Nobel laureate from your research organization isn’t a rounding error. It’s a headline.

Andrej Karpathy, announced just last week, is joining Anthropic in a pretraining research capacity. Karpathy is perhaps the most widely respected AI educator and research communicator of his generation — his YouTube lectures have trained more ML engineers than most university programs — and his institutional knowledge of large-scale model training is rare.

Jonas Adler and Alexander Pritzel are the latest additions. Both were central contributors to Gemini’s model architecture — the system that powers Google Search’s AI features, Workspace integrations, Google Cloud’s enterprise AI products, and the competitive intelligence at the heart of Google’s AI strategy. Their deep institutional knowledge of how Gemini was designed, trained, and deployed is precisely the kind of expertise that cannot be documented in a wiki or transferred to a successor in a month.

Why Are They Leaving?

No single factor fully explains the exodus, but several converging dynamics are visible.

Pre-IPO equity is extraordinarily compelling right now. Anthropic raised a $65 billion Series H in May 2026 at a valuation approaching $1 trillion. OpenAI is preparing a public offering for later in 2026 at a private valuation north of $800 billion. For a researcher making the leap now, the equity on offer at either company represents potential generational wealth. Google is already a public company; the delta between Google RSUs and pre-IPO shares at Anthropic or OpenAI is enormous.

“It’s a great time for the companies to recruit top AI talent with a promise of equity,” TechCrunch noted in its reporting on the departures.

Research culture has shifted at Google. Multiple sources suggest that as the company has moved from treating AI as a research discipline to treating it as a product imperative, the culture of what it means to do research has changed. The pressure to ship competitive Gemini updates on a timeline set by rivals — rather than the timeline set by scientific readiness — has altered the working environment in ways that some researchers have found increasingly difficult to reconcile with why they joined.

Mission clarity at Anthropic and OpenAI. Both companies have built organizational identities around an explicit sense of urgency about AI’s trajectory — not just as a product category, but as a civilization-scale challenge. Working on that problem at a dedicated AI lab, where it’s the entire company’s reason for existing, feels different from working on it inside a $2 trillion conglomerate where AI is one priority among dozens.

Cluster dynamics. Talent at the frontier of AI research is concentrated in small networks. When multiple people with overlapping professional relationships begin leaving an organization, it creates a permission structure — a signal to others that departure is a legitimate choice that serious people make. The departures of Shazeer, Jumper, Karpathy, Adler, and Pritzel didn’t happen in isolation. Each probably made the next one easier.

What It Means for Google

The strategic impact is real, though it’s important not to overstate it. Google still employs thousands of exceptional AI researchers. Gemini 3.5 and its successors are products of systems, pipelines, data infrastructure, and TPU compute advantages that don’t disappear when five people leave. Google’s position in AI remains formidable.

But frontier AI research is not uniformly distributed. Some individuals carry disproportionate knowledge, intuition, and capability. Losing Shazeer, Jumper, Karpathy, Adler, and Pritzel in the same month means losing a concentrated cluster of exactly those people — the ones who have internalized, at a visceral level, what it takes to push the frontier. That’s not replicable at pace, regardless of budget.

The harder problem for Google is what this pattern communicates to researchers still inside the company. Every departure validates the option for others. An organization’s culture is partly determined by the choices its most respected people visibly make. If the visible choice of Google’s most celebrated AI researchers is to leave, the culture message is hard to avoid.

Anthropic’s Calculated Recruitment Push

What’s notable is that Anthropic appears to be executing a deliberate and aggressive recruitment campaign, having pulled at least three prominent researchers from Google in recent weeks alone. The company is scaling its research team as it prepares for public markets, and it has identified Google — logically, given the density of elite AI talent there — as its primary talent source.

This isn’t accidental. Anthropic was itself founded largely by former OpenAI employees, and it has maintained a culture that emphasizes research rigor and safety-focused work in ways that have historically attracted researchers who feel a strong sense of personal mission. The company has invested heavily in making the equity proposition concrete in the current fundraising climate.

OpenAI, meanwhile, has secured Shazeer — arguably the most significant individual acquisition in the talent war so far, given the depth and duration of his impact on the field.

Google’s Response

Google has not publicly addressed the specific departures. Its official position in AI remains extraordinarily strong: it operates the only AI search product at genuine global scale, has TPU compute infrastructure advantages that no competitor can replicate on a multi-year timeframe, and retains enormous bench strength across DeepMind and Google Brain.

But the best talent has options in a way it hasn’t had before, and the companies best positioned to make those options concrete — Anthropic and OpenAI — are doing so aggressively. The frontier of AI is being defined by the people doing the work, not only by the company with the most resources.

Google has the resources. Whether it still has the researchers who will build its most important next chapter is a question June 2026 has made suddenly harder to answer.