On Wednesday, as the workday ended in London, Google CEO Sundar Pichai and Google DeepMind CEO Demis Hassabis delivered news to employees.

Hassabis, DeepMind’s co-founder and long-time steward, would give up the CEO title and move into a chairman role, they said. Minutes later, word emerged that long-time chief scientist Jeff Dean and several other prominent researchers were leaving the company.

Employees in DeepMind’s London office were stunned. They wanted to know whether Hassabis’s move was a promotion or a demotion, whether he would still have meaningful influence over DeepMind’s direction, and whether Google was completing a power shift from the London lab to its Mountain View headquarters.

One DeepMind engineer familiar with the internal reaction told Fortune that employees were “very mixed” about the reshuffle, with some wondering whether the chairman role would eventually leave Hassabis phased out of the company entirely.

In reality, the high-profile exits were only the surface of a difficult year for Google DeepMind. Three current employees and one former employee told Fortune that tensions inside the lab had already been rising. Delayed models, major departures, and a public dispute with staff over controversial defense contracts had left many insiders questioning whether Google had lost its footing in the AI race.

Once considered a leader, the company has recently ceded ground to rivals, especially OpenAI and Anthropic. That shift has been most visible in the repeated delays to Gemini 3.5 Pro, which has missed three release deadlines in recent months. Three DeepMind engineers who spoke to Fortune blamed the delays on the company’s failure to prioritize AI coding capabilities.

For the second time in just four years, Google is again trying to catch up.

Google DeepMind is now struggling to retain its technical edge, keep its top researchers, and manage an employee revolt over defense work. Pichai’s reshuffle, which employees say shifts power back to Mountain View and pulls DeepMind closer to Google proper, comes at a critical moment. It is also a gamble: can new leadership restore Google’s AI momentum, or will it deepen internal tensions during a crucial period in the AI race?

A leadership shakeup

In a note to staff on Wednesday, Hassabis said he was handing over day-to-day operational responsibilities so he could have the “time and space to focus on the big picture and help influence what is to come to the best of my ability.”

But rather than appointing a new DeepMind CEO, Pichai said DeepMind chief technology officer Koray Kavukcuoglu would take over day-to-day operations as senior vice president, reporting directly to Pichai rather than holding a standalone CEO title. Unlike Hassabis, who is based in London, Kavukcuoglu works in Mountain View, about 6,000 miles away. A Google spokesperson said Hassabis was moving into an expanded role focused on advancing science and long-term strategy.

Three DeepMind engineers who requested anonymity to discuss company matters told Fortune they saw the change as part of a broader structural shift, one that moves power toward Mountain View, where Google is headquartered, and away from London, where DeepMind was founded.

They said that losing Hassabis as CEO without naming a replacement also means the lab is being partially absorbed into the wider company and losing some of its independence. The first engineer called it a “small step, but gradually continuing the trend.”

“A lot of people are quite upset by it,” that engineer said. “Both Demis exiting, and losing some of the separation from Alphabet and identity as DeepMind.”

For months, there had been rumors of a larger power shift to California, along with concerns about what that would signal to London-based staff. Workers described ongoing “background pressure” as functions moved to Mountain View, and several prominent London-based engineers, including AlphaFold veterans John Jumper, Jonas Adler and Alexander Pritzel, recently left for Anthropic.

According to the second engineer, reporting lines had also been slowly moving away from Hassabis for some time.

A Google spokesperson rejected the idea that power was shifting to Mountain View and said Google DeepMind remained committed to its London presence. The spokesperson said Hassabis, along with much of Google DeepMind’s leadership, would remain based in London, and that DeepMind would retain separate leadership and research autonomy under Kavukcuoglu.

From a business perspective, there may be an argument for change, especially if the company is struggling to ship models in a more competitive environment. Mountain View may have run out of patience and want tighter control over its AI crown jewel.

“It’s very clear that they’re having a big internal debacle,” the second engineer said. “This might have also contributed to speeding up some of these power shifts. Generally in Google, when something is significantly underperforming or not working, their solution is to just do a huge reorg, a reallocation of compute and priorities to try and fix it.”

DeepMind’s journey

This is not the first time Google has stumbled on its AI path.

In late 2022, the company declared a “code-red” after the release of OpenAI’s ChatGPT, fearing the impact on its core search business. Since then, its catch-up effort has been partly successful. Gemini 3 outperformed rivals and pushed OpenAI to declare its own “code-red” in late 2025, as Gemini’s benchmark gains and rapid user growth put pressure on ChatGPT’s lead.

But the direction has since changed.

Google’s models are now not competitive with the leading edge of the U.S. frontier on intelligence and coding benchmarks. Even some open-source Chinese efforts are outperforming the tech giant.

Gemini 3.5 Pro was meant to be Google’s next major response to advanced models such as Anthropic’s Mythos and OpenAI’s GPT 5.6. Introduced at the company’s I/O conference in May with a promise to ship “next month,” it missed the June target and then another mid-July date. By August, the model was still unreleased, even as Google shipped three smaller Flash models around it and said it had already begun training Gemini 4.

Micah Hill-Smith, co-founder and CEO of benchmarking firm Artificial Analysis, told Fortune that Gemini 3.6 Flash, Google’s strongest shipped model, now ranks behind models from Anthropic, OpenAI, one or two leading Chinese labs, xAI and Meta on his company’s intelligence index, which measures overall AI model performance.

“For a brief moment, we actually did have [Google’s] model on top as the smartest model in the world. Right now, it’s fair to say that [model] from Google is less compelling,” he said.

According to a third engineer, many of the mistakes that led to this began last summer, when the company failed to prioritize Gemini’s coding abilities. Google, the engineer said, was focused on protecting search traffic and beating OpenAI on chat, while rivals surged ahead in agentic and coding capabilities. Hill-Smith said Google’s models now significantly lag rivals on coding-agent benchmarks.

A Google spokesperson said Gemini 3.5 Pro was currently in testing and that the DeepMind team was already working on the next generation of models. The spokesperson acknowledged that Google needs to improve its coding and agentic capabilities, but said the company’s teams were focused on that work.

Google may also have underestimated how central coding would become in consumer products just a year later. Developer tools such as Claude Code and OpenAI’s Codex are now helping drive a new wave of consumer and general-use products. Those include Anthropic’s Cowork, an agentic tool built on the same underlying technology as Claude Code that allows non-developers to hand off multi-step knowledge work such as organizing files or drafting documents, and OpenAI’s superapp, positioned as a one-stop entry point for AI tools.

“They dropped the ball a little bit [on coding],” the third DeepMind engineer said. “By the time you realize something like that…a number of months have passed, and now you have to play catch-up. Then it also becomes political.”

Coding is now a “big focus” inside the lab, the second engineer added.

Still, falling behind on pure model performance does not mean Google is out of the AI race. The company has other strengths, including vertical integration across applications, cloud and chips, which can offset weaker performance at the frontier of raw model intelligence.

“For OpenAI and Anthropic, because the model layer is so important to what they do, it’s pretty existential for them to win there,” Hill-Smith said. “Google could end up doing very well in the AI race overall, even if they’re not at the frontier of the model layer,” he said, adding that Google’s unusually deep vertical integration is a structural advantage.

Falling behind — and jumping ship

The technical struggles and the sense of having to catch up twice in less than five years have hurt morale, according to all the current and former DeepMind workers who spoke to Fortune.

That frustration has overlapped with one of the most damaging waves of departures in DeepMind’s history. In one week in June, Google lost Gemini co-lead Noam Shazeer to OpenAI and Nobel laureate and AlphaFold co-inventor John Jumper to Anthropic.

The employees who spoke to Fortune said those exits were driven by overlapping forces: aggressive poaching from well-funded rivals, frustration at being behind in the AI race, and the appeal of joining pre-IPO labs before they go public, where equity compensation could be especially lucrative. “There’s a lot of just hopping over to pre-IPO labs,” one employee said.

A Google spokesperson said talent regularly moves between technology companies and that Google remained focused on attracting people aligned with DeepMind’s mission.

Google has tried to limit the exodus. The second engineer said the company has deployed large retention and equity bonuses for some staff, building on an earlier round of enhanced compensation during a period when Meta was making aggressive counteroffers to AI researchers.

Employees say those efforts have not been enough to offset a deeper issue: burnout.

“I’ve been pushing 60-hour weeks for quite a long time because there is so much to do,” the third engineer said. “People are pushing themselves really hard because it is a race, and you have this feeling that you need to catch up… [but] telling people to work 60-hour weeks constantly and that it is constant crunch time usually doesn’t work that well, and people do burn out.”

Google said AI is an intensely competitive field and pointed to researchers who have publicly expressed excitement about their work and the company’s direction.

There were also complaints about leadership. Even before his departure, workers described Hassabis as absent, particularly compared with Google co-founder Sergey Brin, who returned from retirement to help with the company’s AI efforts.

“You will see Sergey being incredibly hands-on, working hard on stuff and being involved,” the third DeepMind engineer said. “I can’t say that I’ve ever seen Demis walk around the office in the Gemini area.”

“Demis isn’t a very inspiring leader,” the former employee added. “Certainly not the way people talk about Dario [Amodei] … Demis is more of a guy who is known but not seen.”

A union battle and contentious defense deals

The leadership and talent concerns have collided with a separate, more politically charged dispute over Google’s defense business.

In April 2026, Google signed a deal with the Pentagon allowing the Department of Defense to use Gemini models on classified networks for any lawful government purpose. More than 580 Google employees, including senior DeepMind researchers, later signed an open letter urging Pichai to reconsider the deal, citing concerns about autonomous weapons and mass surveillance.

Google was not alone in agreeing to such a deal. OpenAI, xAI, Microsoft, Amazon and others signed similar classified-use agreements. Anthropic was the outlier, refusing to drop its restrictions on autonomous weapons and mass surveillance. That position led the Pentagon to label Anthropic a supply-chain risk and order the military to stop using its products.

At Google, the Pentagon agreement reopened a longer-running fight over defense and intelligence contracts. Employees have also objected to Project Nimbus, the roughly $1.2 billion cloud-computing deal Google and Amazon signed with the Israeli government in 2021.

For many employees, the Pentagon agreement slipped under the radar. For others, it was existential.

Alex Turner, who spent more than two years on DeepMind’s AI safety team, resigned in June and said the Pentagon deal was the trigger. In a public post, he said months of internal advocacy had gone nowhere, and that he had even tried to enlist Google chief scientist Dean as an ally.

Andreas Kirsch, another DeepMind research scientist, was also a vocal public critic. He posted that the contract restrictions amounted to little more than “meaningless weasel words” for public relations purposes, and said he felt “incredibly ashamed” to be associated with the lab over the decision.

The Pentagon deal became the final straw for a group of London-based DeepMinders, who launched a union drive in May in an attempt to regain some leverage. It is the first formal effort to unionize at a frontier AI lab.

The campaign traces back to February 2025, when Google removed language from its public AI Principles that had committed the company not to build AI for weapons or surveillance that violate international norms. Many engineers were angered by what they saw as a retreat from Google’s ethical assurances. The company had adopted that pledge in 2018 after the backlash over Project Maven, a separate defense contract.

Google has declined to voluntarily recognize the union, instead offering to talk through the UK conciliation service ACAS. The two sides have already clashed over basic issues, including which employees should be included in the bargaining unit, according to a person familiar with the situation. Employees now worry that American leadership may be less willing to compromise in the negotiations.

A Google spokesperson said the company was still in the early stages of the union process and would continue to negotiate in good faith.

DeepMind’s uncertain future

With Hassabis stepping back and Dean gone, DeepMind appears to be entering a new phase at a moment when it can least afford disruption.

The lab still has advantages that few rivals can match: distribution, a cloud business, custom chips, a large research bench and the ability to embed AI across products used by billions. But the reorganization appears to remove a layer between DeepMind and a parent company increasingly focused on turning AI research into commercial and government deployments.

For employees, the question is whether the new leadership structure and tighter integration will bring the urgency and clearer direction needed to close the gap in coding and agents, or worsen the problems already pushing people out: burnout, reduced autonomy and concerns about how the company will use its technology.

Just four years after Google’s first AI “code red,” DeepMind is again trying to prove it can move quickly enough to keep pace. This time, it will do so without the two figures most closely associated with its scientific identity.

This story was originally featured on Fortune.com