Demis Hassabis is stepping down as chief executive of Google DeepMind to become chairman of the AI lab, while longtime Google chief scientist Jeff Dean is leaving the company after 27 years to co-found a startup, according to an internal announcement on Wednesday that sent Alphabet shares down roughly 5 percent.
The reshuffle, which current and former employees told Fortune had been preceded by months of internal tension, arrives as DeepMind struggles with repeated delays to its Gemini 3.5 Pro model, an accelerating departure of senior researchers, and an employee revolt over a Pentagon defense contract that has already triggered resignations and a union drive.
News Summary
- Demis Hassabis stepped down as Google DeepMind CEO on August 5, 2026, becoming chairman and chief scientist of Alphabet. He retains leadership of Isomorphic Labs.
- Jeff Dean, a 27-year Google veteran and chief scientist, is leaving to co-found Discovery Loop with Sanjay Ghemawat, Oriol Vinyals, and Quoc Le. Alphabet will be a founding investor and cloud partner.
- DeepMind CTO Koray Kavukcuoglu is taking over day-to-day operations as senior vice president, reporting directly to Sundar Pichai from Mountain View.
- DeepMind engineers told Fortune the move shifts power from London to Mountain View and reduces the lab’s independence.
- Gemini 3.5 Pro has missed three release deadlines. Independent benchmarks place Google’s strongest shipped model, Gemini 3.6 Flash, behind offerings from Anthropic, OpenAI, xAI, Meta, and some Chinese labs.
- High-profile recent departures include Gemini co-lead Noam Shazeer to OpenAI and Nobel laureate John Jumper to Anthropic.
- More than 580 employees signed an open letter opposing a Pentagon contract signed in April 2026. DeepMind AI safety researcher Alex Turner resigned in June over the deal.
- London-based DeepMind staff launched a union bid in May, the first formal attempt at a frontier AI lab.
The Leadership Shake-Up
In a note to staff, Hassabis wrote that he was handing over operational responsibilities to gain “time and space to focus on the big picture and help influence what is to come to the best of my ability.” He also took on the title of chief scientist at Alphabet and will continue leading Isomorphic Labs, the company’s AI drug-discovery unit.
Rather than appoint a new CEO, Google placed DeepMind CTO Koray Kavukcuoglu in charge of day-to-day operations as senior vice president, reporting directly to Pichai. Kavukcuoglu works from Mountain View, roughly 6,000 miles from DeepMind’s London headquarters. A Google spokesperson told Fortune that Hassabis and much of DeepMind’s leadership would remain based in London, and that the lab would retain separate leadership and research autonomy.
Three current DeepMind engineers, speaking anonymously to Fortune, described the change differently. They said it represents a structural shift of power to Mountain View and a gradual absorption of DeepMind into the broader company. One engineer called it a “small step, but gradually continuing the trend,” adding that “a lot of people are quite upset by it.”
The second engineer told Fortune that Google had been slowly moving reporting lines away from Hassabis for some time, and that rumors of a significant power shift to California had circulated for months. Several prominent London-based researchers, including AlphaFold veterans John Jumper, Jonas Adler, and Alexander Pritzel, recently left for Anthropic.
Model Delays and Benchmark Slippage
The personnel changes come as DeepMind’s flagship Gemini 3.5 Pro model remains unreleased after missing three deadlines, including a June target and a mid-July date. The model was unveiled at Google’s I/O conference in May with a promise to ship “next month.”
Three DeepMind engineers who spoke to Fortune blamed the delays on the company’s failure to prioritize AI coding abilities last summer, while rivals were advancing on agentic and coding capabilities. Google, they said, was preoccupied with protecting search traffic and competing on chat, allowing competitors to pull ahead.
Micah Hill-Smith, cofounder and CEO of independent benchmarking firm Artificial Analysis, told Fortune that Gemini 3.6 Flash , the strongest model Google has shipped , now sits behind models from Anthropic, OpenAI, one or two leading Chinese labs, xAI, and Meta on his firm’s intelligence index. “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.
A Google spokesperson told Fortune that Gemini 3.5 Pro was in testing and that the team was already working on its next generation of models. The spokesperson acknowledged that the company needed to improve in coding and agentic capabilities, but said its teams were focused on that work.
Burnout and the Talent Exodus
The technical setbacks have coincided with what current and former employees described to Fortune as one of the most damaging stretches of departures in DeepMind’s history. In a single week in June, Google lost Gemini co-lead Noam Shazeer to OpenAI and Nobel laureate John Jumper to Anthropic.
Employees told Fortune the exits are driven by overlapping forces: aggressive poaching from well-funded rivals, frustration at falling behind in the AI race, and engineers seeking equity in pre-IPO labs. “There’s a lot of just hopping over to pre-IPO labs,” one employee said.
Burnout appears to be compounding the problem. One engineer told Fortune they had been working 60-hour weeks for an extended period. “People are pushing themselves really hard because it is a race, and you have this feeling that you need to catch up,” the engineer said, adding that constant crunch time “usually doesn’t work that well, and people do burn out.”
Google has deployed retention and equity bonuses for some staff, according to one engineer, but employees said the compensation has not been enough to offset the deeper morale problem. Workers also described Hassabis as absent from day-to-day Gemini operations, especially compared to Google cofounder Sergey Brin, who returned from retirement to assist with AI efforts. “I can’t say that I’ve ever seen Demis walk around the office in the Gemini area,” one engineer told Fortune.
Defense Contracts and Employee Revolt
The internal frustration over leadership and model delays has collided with a politically charged fight over Google’s defense work. In April 2026, Google signed a deal allowing the Pentagon 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, citing concerns about autonomous weapons and mass surveillance.
Alex Turner, who spent more than two years on DeepMind’s AI safety team, resigned in June and cited the Pentagon deal as the trigger. In a public post, he said months of internal advocacy had gone nowhere. Andreas Kirsch, another DeepMind research scientist, publicly called the contract’s restrictions “meaningless weasel words” and said he felt “incredibly ashamed” to be associated with the lab.
The controversy 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 violating international norms. That pledge was originally adopted in 2018 after backlash over Project Maven, a separate defense contract. Employees have also criticized Project Nimbus, a roughly $1.2 billion cloud-computing agreement Google and Amazon struck with the Israeli government in 2021.
In May, a group of London-based DeepMind staff launched a union bid, the first formal attempt by a frontier AI lab to unionize. Google has declined to voluntarily recognize the union, offering talks through the U.K.’s conciliation service, ACAS, instead. A person familiar with the situation told Fortune the two sides have already clashed over which staff should be included in the bargaining unit.
What Comes Next
With Hassabis stepping back and Dean gone, DeepMind is entering a new phase at a moment when it can least afford further disruption. The lab still retains structural advantages , distribution across billions of users, cloud infrastructure, custom chips, and vast research resources , that could offset weaker performance at the frontier model layer, according to Hill-Smith. “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 told Fortune.
For employees, the immediate question is whether tighter integration with Mountain View will bring the urgency and clearer direction needed to close the gap in coding and agents, or deepen the problems already driving people out. One engineer told Fortune that when something at Google 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.”
Four years after Google’s first AI code red following the launch of ChatGPT, DeepMind is again trying to prove it can move fast enough to keep pace. This time, it will do so without the two figures most closely associated with its scientific identity.