Jonas Adler and Alexander Pritzel, both seen internally as central contributors to Google’s Gemini AI model, are expected to move to the company that created Claude, said the sources, who requested anonymity because the information is not public. Adler worked on the company’s AI programming effort, and Pritzel worked on the artificial intelligence systems training process.
Google, an early pioneer in artificial intelligence, spent much of the current AI boom chasing companies like OpenAI and Anthropic before gaining traction late last year with more advanced models and chips.
In recent days, however, the company had already lost two prominent names: Nobel winner John Jumper, who goes to Anthropic, and star researcher Noam Shazeer, who went to OpenAI. The moves rattled investors and cast new doubts on Google’s ability to compete in the fierce race to build better models.
InfoMoney Tool
Download now (and for free)!
The departures highlight the pressure Google faces from two startups that are close to going public, even offering highly paid Big Tech employees the chance for a rare gain by joining before an IPO. In at least one case, a Google professional’s departure also appears to have been preceded by changes in priorities about how to allocate scarce computing resources, an issue that led other employees to leave the company altogether.
Shortly before Shazeer announced its plans to join OpenAI, computing capacity earmarked for one of its projects was reallocated to a London-based team at Google DeepMind, according to two people familiar with the matter. The change was made in an attempt to increase collaboration between teams and simplify Google’s work in pre-training, the initial phase of AI development in which models learn from large volumes of data, the sources said.
Adler, Pritzel, Jumper and Shazeer did not respond to requests for comment. Anthropic declined to comment. A Google spokesperson said the company remains confident in its position in the AI talent market and cited statements Google DeepMind CEO Demis Hassabis made earlier in the week.
“There is a lot of talent movement among all the major labs, and we get a fair share of the best talent. We have by far the largest and broadest pool of researchers among all the labs,” Hassabis said at an event in Cannes. “This is a fiercely competitive market right now, the most fiercely competitive it’s ever been in the technology industry.”
Shazeer’s career path is emblematic of the intense wars for talent that have come to define the AI landscape. After co-signing a seminal paper that helped catalyze the artificial intelligence boom, he left Google in 2021 to found chatbot startup Character.AI, only to return to the company in 2024 as part of an unusual licensing deal that valued his company at $2.5 billion.
Back at Google, Shazeer began co-directing the development of the company’s main AI model, Gemini. Before his departure, he had also been working on a new AI architecture, two people said. The architecture was still based on the transformer, a technique introduced by Shazeer and his colleagues in 2017 and which became a pillar of AI development in the following years, but had been showing promising results, according to one of the sources.
Continues after advertising
Shazeer was both an admired and divisive figure within Google, current and former employees say. Comments he made within the company about transgender identity and the conflict in Gaza sparked controversy among some employees, according to two people.
Jumper, meanwhile, had become one of the faces of Google’s most ambitious AI efforts after winning the Nobel Prize for landmark research that used AI to predict protein folding. Adler and Pritzel, both soon to join Jumper at Anthropic, worked with him on this research.
Key members of Jumper’s team on protein work have left Google DeepMind in recent months. Some have migrated to Isomorphic Labs, an Alphabet spin-off company focused on AI-powered medicines, according to a person familiar with the matter.
Continues after advertising
Anthropic, which both competes with Google and maintains a partnership with it, has been aggressively draining talent from the technology giant. DeepMind engineers are nearly 11 times more likely to leave for Anthropic than the other way around, according to a 2025 industry analysis from venture capital manager SignalFire.
Like Google, the company that created Claude also explores applications in life sciences and health, in an attempt to expand the uses of its technology. Anthropic recently raised a new round of financing with a valuation of US$965 billion, surpassing OpenAI, and is considering going public this fall.
AI researchers in the United Kingdom, where DeepMind’s leadership is based, are often subject to lengthy non-compete agreements, which can be enforced under British law. Jumper likely wouldn’t start working at Anthropic until next year, according to a person familiar with the matter.
Continues after advertising
Another researcher, Arthur Conmy, wrote in X on Wednesday that he was also leaving for Anthropic to work on AI security. During his time at DeepMind, Conmy was a senior research engineer and contributed to the Gemini 2.5 model, in addition to working in AI programming, according to his LinkedIn profile.
© 2026 Bloomberg L.P.
