At Google’s I/O 2026, DeepMind CEO Demis Hassabis said the company is pushing toward Artificial General Intelligence while taking on the problems that come with AI showing up in daily products—agents, synthetic media authenticity, cybersecurity, and even futur
On Tuesday. Google’s I/O 2026 keynote ran so packed with announcements that it felt like the show might burst at the seams. For three hours. CEO Sundar Pichai and other executives lined up news across Google Search. the Gemini app. Google Docs. Gmail. YouTube. Android. and more—an AI-heavy update that’s become a yearly benchmark for how close the tech giant thinks it is to the next leap.
For Demis Hassabis, the CEO of Google DeepMind, the pace is personal. His path to these products stretches back to childhood fascination with teaching machines to think. In 2010, that obsession helped lead him, alongside Shane Legg and Mustafa Suleyman, to cofound DeepMind. Google acquired DeepMind in 2014 and merged it with another research arm, Google Brain, in 2023. The work. Hassabis said. is still moving toward one big goal: Artificial General Intelligence—AI that’s at least on par with human thinking across an array of domains.
That destination comes with disagreements even among people inside the field. Google Brain cofounder Andrew Ng believes AGI is decades away. Hassabis, though, says the timeline is closer. “2030 is when I expect it to arrive, either plus or minus a year,” he said.
The reason the question matters isn’t confined to labs anymore. AI is already part of everyday life through Google products, and its promise—and pitfalls—are becoming harder to ignore. When Hassabis sat down to discuss the I/O announcements. he talked about new features with the excitement of someone watching his own work turn into daily tools. He was just as energized by the problems AI can cause, and what Google is doing to mitigate them. He underlined that advancing the science of AI remains “my main passion.”.
He also framed the current moment as a balancing act: “It’s complicated, because you’ve also got the most voracious competition in tech history going on,” he told the interviewer. “I won’t pretend that it’s easy. But I think we get that balance right better than anyone else.”
One of the keynote’s most concrete examples of AI moving from research to product is Gemini Spark. which Hassabis described as part of a longer internal story about “agentic” systems. When asked about the new Gemini agent. Hassabis pointed to DeepMind’s earlier work in agentic AI—saying AlphaGo was an agent and that the team’s original Atari work also operated as agents.
Gemini Spark includes Daily Brief, a summary of a user’s current doings. Hassabis said the aim is to help a wider set of people, not just those with technical aptitude. “The sweet spot is to help everyone with these agents, not [just] people who are very technical,” he said. “But also to make sure it’s actually secure. reliable. and robust. and you have full control over what it has access to.”.
His comments also pointed to the risks he sees in other agent experiments. He referenced OpenClaw, saying running its best-known version requires considerable technical aptitude and a willingness to risk things going awry. Enthusiasts, he said, also conclude it needs a budget big enough to dedicate a Mac Mini to the job. By contrast. Hassabis said Gemini Spark runs 24/7 in the cloud. connects only to apps users explicitly authorize. and for now “just works with other Google services.”.
He was blunt about security concerns around OpenClaw: “One of the main issues with OpenClaw is it’s just very insecure. I wouldn’t recommend it for any real work. I haven’t used it for any of my real stuff, because it might leak everything.”
Gemini Spark is rolling out first to users who subscribe to Google’s high-end $100/month AI Ultra plan. Hassabis said expanding access to more people will mean adapting the agent to average workflows. “Adapting to the average person, adapting their workflows to this type of agentic assistance,” he said. “It’s probably going to play out over the rest of the year, would be my guess.”.
While Gemini Spark focuses on agents that can act across authorized apps. another announcement at I/O—Gemini Omni Flash—targets what happens when models handle more than text. Hassabis singled it out as the announcement he’s most excited about. Gemini Omni Flash can take text. images. video. and audio as part of a prompt. and it will debut in the Gemini app. Google Flow video editor. and YouTube Shorts. where it will output video. Eventually, Hassabis said, it will be able to generate other forms of media.
“What people are going to be able to do is experiment between different modalities,” he said. “Here’s a video input, here’s a music output, here’s an image input. Give me a video output. I just want people to be incredibly creative with it.”
The creativity is arriving alongside a credibility problem the industry knows too well: AI-generated media can look convincing even when it isn’t. Hassabis noted that Google’s Gemini models are increasingly capable of generating realistic imagery. but he said making it easy to determine what’s AI-created matters.
Google’s approach has been evolving for years. The company introduced SynthID in 2023, describing it as identification technology for AI-generated content. Since then. Google said it’s watermarked “over 60 years of video and 100 billion images.” It has also championed C2PA Content Credentials. a standard meant to track whether imagery was created with a camera or AI and how it has been modified.
At I/O. Google moved the effort further into mainstream experiences—building the provenance tools into widely-used platforms such as the Gemini app. Android’s Circle to Search. and Google Search’s AI Mode. The push wasn’t confined to Google either. The keynote included an announcement from OpenAI that it would add support for SynthID in ChatGPT, Codex, and its API.
Hassabis said the momentum is the point. “I think it’s great that the whole industry is coalescing around watermarking that is robust,” he said. “That’s what was needed, really, to then go the next step, which is having sites automatically identify [content authenticity]. Or you can imagine even browsers eventually doing that, so there’s almost no effort in terms of verifying something.”.
From media authenticity, the conversation moved to cybersecurity. Hassabis was asked whether Google is wrestling with the same kinds of issues other AI leaders are confronting, including the risk that AI could help find and exploit software vulnerabilities. He said the answer is yes.
He pointed to Google’s position as an advantage for helping developers secure their apps in the AI era. highlighting assets such as its CodeMender agent and the Wiz cybersecurity platform. as well as noting Wiz as Google’s largest-ever acquisition. Still, Hassabis emphasized that blocking AI from giving “superpowers” to bad actors isn’t the only urgent task.
Over the next year or year and a half. he predicted AI could accelerate chemical. biological. radiological. and nuclear (CBRN) threats. “I’m thinking a lot about what sorts of tools. monitoring systems. and other things all the frontier labs should really be working on and implementing. ” he said. He also pointed to chain of thought monitoring. which lets researchers deconstruct a model’s thought process and look for signs that it’s engaging in deceptive behavior.
“There’s a lot that we’re sort of in the foothills of now,” he said. “Models that are super capable, which is great. And they’re agentic, also great. But that means there are more challenges and risks associated with them.”
Beneath the caution is a more optimistic thesis: Hassabis wants AI to become “the ultimate tool to solve all the world’s most complex scientific problems. ” quoting the voiceover at the start of the I/O keynote. That belief is showing up in how he spends his time. Along with running Google DeepMind. he also leads Isomorphic Labs. a spin-off focused on commercializing its AlphaFold protein stricture prediction AI for use in drug discovery.
AlphaFold’s progress is already drawing major attention. Hassabis and John Jumper, Distinguished Scientist at Google DeepMind, shared the 2024 Nobel Prize in Chemistry for AlphaFold’s creation. Last week, Isomorphic Labs announced that it raised $2.1 billion in new funding. Hassabis said it’s a strong sign of confidence: “You can take that as a huge vote of confidence in the progress we’re making over there.”.
As AlphaFold edges closer to real-world impact, other DeepMind projects with long-term ambition are also moving. Hassabis said the company is collaborating with the U.K. government to build an automated science lab that uses the Gemini LLM and robotics to investigate areas such as superconducting materials and nuclear fusion.
He returned repeatedly to the constraints that make these efforts feel endless. “Obviously, there’s never enough compute for the ideas that you have,” he said. “But I think we’ve done that historically very well at DeepMind, originally, and now Google DeepMind—just protecting blue-sky research.”
Even during I/O week—when the product announcements provide evidence that Google knows how to deploy AI—Hassabis’ attention stayed forward. His message wasn’t just that AI is moving fast. It was that the hardest questions are moving with it.
Demis Hassabis Google I/O 2026 Sundar Pichai DeepMind Artificial General Intelligence AGI Gemini Spark Gemini Omni Flash SynthID C2PA Content Credentials OpenClaw CodeMender Wiz AlphaFold Isomorphic Labs $2.1 billion funding