On 5 August 2026 Alphabet reshuffled Google DeepMind. Demis Hassabis stepped down as chief executive, and Koray Kavukcuoglu was promoted to senior vice president, Reuters reported and The Guardian reported. The standalone chief executive role was not replaced, and DeepMind’s non-technical teams were moved into Google’s corporate structure. That is the structural fact beneath the headline.
This article traces what changed, who Kavukcuoglu is, why Hassabis stepped back, the Gemini 3.5 Pro delay, and what a product-first pivot means for the model roadmap. It is one part of the full DeepMind leadership-transition picture.
What follows leans on what Google, Reuters and other outlets reported, and flags what remains unconfirmed.
What Exactly Changed in Google’s AI Leadership on 5 August 2026? #
The short answer: the standalone DeepMind chief executive role disappeared. Hassabis became chairman of Google DeepMind and took a new title as Alphabet chief scientist, while Kavukcuoglu moved from chief technology officer to senior vice president, reporting directly to Sundar Pichai. Alphabet announced the moves and Reuters confirmed them. Alphabet did not explain what prompted them. At a 6 August all-hands meeting, staff were told several non-technical teams, including safety, ethics and policy, were being moved out of DeepMind and into Google’s corporate reporting structure, Time reported. That is the part that signals a structural change.
The difference from 2023 matters. The Google Brain merger consolidated two labs under Hassabis as chief executive. This reshuffle subordinates DeepMind under corporate Google. The chief executive title was replaced by an SVP report, and that line to Pichai removes the buffer that once sat between research and Alphabet’s quarterly pressure. Kavukcuoglu now works from Mountain View, not London, making the power shift geographic as well as structural. A former Google executive told The Guardian that “the era of DeepMind as an independent actor is over”. If you read reporting-line changes as vendor risk, this one points the same way.
Why Did Demis Hassabis Step Down as CEO of Google DeepMind? #
Hassabis framed the move as a way to focus on the big picture, telling staff he wanted time and space to think about what comes next, The Guardian reported. His new Alphabet chief scientist role is advisory, aimed at the societal impacts of AGI, long-horizon science and safety governance, CNBC reported. He keeps the DeepMind chairman role and stays chief executive of Isomorphic Labs.
What he has surrendered is day-to-day operational control of Gemini and the DeepMind product roadmap. The timing points to the cause. The move landed amid low morale and staff burnout, Fortune reported via Yahoo Finance, against the backdrop of the model delays covered in the next section. Fortune’s reporting describes a wider pattern of strain; one engineer’s account of 60-hour weeks and crunch time, saying people “do burn out”, is a single anecdote within it. Reports differ on whether he was pushed. Time reported Semafor’s account that he was not, and that he had been handing Gemini responsibility to Kavukcuoglu for a year. Either way, the reporting points to a structural exit.
What Is the Gemini 3.5 Pro Delay, and Why Does It Matter? #
Gemini 3.5 Pro was unveiled at I/O in May 2026 with a promise to ship the next month. It missed three consecutive release deadlines and slipped past its planned June launch. By August Google had delayed the flagship model by a further two months after internal testing showed it still lagged rivals on coding, Reuters reported. Google has not confirmed a replacement date.
The delay cost Google at the frontier. OpenAI’s GPT-5.6 and Anthropic’s Claude Mythos pulled ahead on coding and benchmark performance, and Morningstar analyst Malik Ahmed Khan said Google lost out on coding, “the first clear killer use case”, CNBC reported.
Why did a single model slip become a leadership event? Gemini is Google’s flagship proof of frontier competitiveness, and the delay exposed friction in the old structure. TPU allocation was split across DeepMind, Cloud customers and consumer products, and multiple Gemini project leads disagreed on where to focus. The market reaction showed the link: Alphabet shares fell 4% on the announcement, shedding roughly $250 billion in value in a single session, Reuters reported. The delays also accelerated the talent exodus covered elsewhere in this cluster.
Who Is Koray Kavukcuoglu, Google DeepMind’s New SVP? #
That delay is the reason an operator now runs the lab. Kavukcuoglu is DeepMind’s old guard turned operator. He joined in 2012, before the Google acquisition, and his research background is in reinforcement learning, including deep Q-network work that learned to play Atari games, according to his published bio. He later served as DeepMind’s chief technology officer and Alphabet’s chief AI architect.
The contrast with Hassabis is the point. Hassabis’s remit was research identity and long-horizon science; Kavukcuoglu’s is execution. Sebastian Mallaby has described Hassabis as Google’s AI statesman while Kavukcuoglu ran the research, Business Insider reported. Kavukcuoglu now oversees Gemini model development, Frontier AI research, and the Gemini app and developer teams. He also has final say on major DeepMind decisions.
Day to day, that consolidates power in Mountain View and sits him close to Google Cloud on TPU allocation, with Sergey Brin’s backing, 36Kr reported. The appointment signals a goal of shipping faster and closing that coding gap through product integration.
What Does a Product-First Shift Mean for Google’s AI Model Roadmap? #
Under an execution-first SVP, the roadmap now runs on revenue relevance: DeepMind’s research increasingly serves shipping Gemini features across Search, Cloud, Android and YouTube, and sequencing follows enterprise demand more than long-horizon science.
Long-horizon AGI and moonshot work gets deferred or externalised. Discovery Loop, the public benefit corporation founded by Jeff Dean and three other departing AI leaders, is one example: it is backed by outside venture capital while Google keeps a founding stake and its position as exclusive cloud partner, according to an analysis of the shake-up. Google retains control of the TPU infrastructure those efforts depend on, so whatever Discovery Loop achieves, the chips stay Google’s. Brin has been informally pushing resources toward recursive self-improvement, the point where AI improves itself without human intervention, now folded inside product goals, Reuters reported.
If your business is evaluating Gemini Enterprise or Google Cloud, the intended upside is faster delivery; the trade-off is that oversight teams have moved into corporate Google. That is the talent exodus and research sacrifice behind the pivot, and what the restructuring means for enterprise buyers.
What It Actually Changes #
The reporting line carried the change. An SVP reporting directly to Pichai is the mechanism by which Alphabet’s quarterly product pressure now reaches the lab without a research buffer. Once you see that, the next Google AI story, a model delay, a spin-out, a compute decision, reads as downstream of this structural shift, not an isolated event.
The open question is whether subordinating research to Gemini shipping actually closes the frontier gap, or just defers the long-horizon AGI work DeepMind existed to do. That question will shape the next few quarters of this leadership transition more than any org chart.
Frequently Asked Questions #
Is Demis Hassabis leaving Google and Alphabet altogether?
No. Hassabis remains inside Alphabet as chairman of Google DeepMind and as Alphabet chief scientist, a role focused on AGI’s societal implications, long-horizon science and safety governance. He also stays CEO of Isomorphic Labs. What he has given up is day-to-day operational control of Gemini and the DeepMind product roadmap, not his association with the company.
Has Google DeepMind been absorbed into Google and wound down?
No. The lab still exists and keeps its research identity under the Google DeepMind name. What changed is the reporting line: the standalone CEO role was replaced by an SVP who reports directly to Sundar Pichai, and non-technical teams such as safety, ethics and policy were moved into corporate Google. That narrows the lab’s autonomy, but it does not dissolve it.
Does the reshuffle weaken Google’s AI safety and ethics oversight?
The non-technical teams that handled safety, ethics and policy were moved out of DeepMind’s independent hierarchy and back into corporate Google. That does not automatically weaken oversight, but it changes where those concerns are weighed: closer to Alphabet’s product and commercial structures than to the research lab. For critics, that is the clearest sign the reshuffle was structural rather than a routine title change.
Does this mean Google is giving up on AGI and long-horizon research?
Not explicitly, but the centre of gravity has moved. AGI and moonshot work is being deferred or externalised rather than cancelled, and Sergey Brin’s recursive self-improvement bet now sits inside product goals. Hassabis keeps an advisory remit on AGI safety as chief scientist. The real risk is that long-horizon research becomes the first thing deprioritised when quarterly product pressure rises.
When is Gemini 3.5 Pro actually going to ship?
Google has not confirmed a new release date. The model missed three consecutive deadlines and slipped past its planned June 2026 launch, which is what turned the delay into a leadership event. Under Koray Kavukcuoglu’s execution-first remit, the stated aim is a faster product cadence, but no public timeline has replaced the missed one.
What does this mean for businesses buying Google Cloud or Gemini Enterprise?
For enterprise buyers, the reshuffle is best read as a vendor-stability signal rather than a product guarantee. A direct reporting line to Pichai and an execution-focused SVP are meant to speed up Gemini delivery and close the coding gap with OpenAI and Anthropic. The counterweight is that oversight teams have moved into corporate Google, which changes how safety and policy decisions are made.
What is Isomorphic Labs, and why is Hassabis staying there?
Isomorphic Labs is Hassabis’s drug-discovery company, which applies AI to biology. Keeping the CEO role there while stepping back from DeepMind operations fits the pattern: he is retreating to long-horizon science and its applications while Alphabet puts an operator in charge of shipping Gemini. It also signals his exit was about DeepMind’s product pressure, not a clean break from AI research.
What is a TPU, and why does TPU allocation matter here?
A TPU, or tensor processing unit, is Google’s custom chip for training and running AI models. Allocation matters because Gemini depends on scarce TPU capacity, and tension over who gets that compute exposed friction in the old structure. Part of Kavukcuoglu’s new remit is liaising with Google Cloud on TPU allocation so product delivery is not held up by compute bottlenecks.
Will users notice changes in Search, Android or YouTube?
Not immediately, but the direction points that way. A product-first shift means DeepMind research increasingly serves shipping Gemini features across Search, Cloud, Android and YouTube, sequenced by revenue relevance and enterprise demand. Users are most likely to notice faster Gemini feature rollouts in Google’s consumer products, with less attention going to long-horizon research as the trade-off.
Is Google permanently behind OpenAI and Anthropic on coding?
The gap is real but not necessarily permanent. Gemini 3.5 Pro’s delay let OpenAI’s GPT-5.6 and Anthropic’s Claude Mythos pull ahead on coding and frontier benchmarks. Kavukcuoglu’s appointment is explicitly aimed at closing that gap through faster product integration. Whether execution alone can do it, without the long-horizon research DeepMind was built for, is the open question the restructuring leaves unanswered.
What is the Discovery Loop spin-out?
Discovery Loop is one example of Google externalising long-horizon work rather than carrying it on the core balance sheet. In the product-first model, moonshot and AGI research gets deferred or pushed into separate vehicles while Google retains control of the TPU compute those efforts depend on. It is part of the compute-landlord pattern: keep the chips, externalise the riskier science.