Google DeepMind

What Google DeepMind does

Google DeepMind is the artificial intelligence division of Google, whose parent company is Alphabet. It was formed on April 20, 2023, when Google combined DeepMind with the Brain team from Google Research. DeepMind itself started in 2010 as a London lab, and Google agreed to buy it in January 2014. The unit develops the Gemini models behind Google's chatbot app and developer tools, as well as models for images, video, music and robotics, according to its own overview. It also builds scientific systems, including AlphaFold, which predicts protein structures, AlphaGenome, which predicts the effects of DNA variants, and the WeatherNext forecasting models.

Why it matters

Gemini competes with OpenAI's ChatGPT and Anthropic's Claude. Google chief executive Sundar Pichai said in August 2026 that the Gemini app had passed 1 billion monthly active users. Half of the 2024 Nobel Prize in Chemistry went jointly to Demis Hassabis and John Jumper, both then at Google DeepMind, for protein structure prediction. The Royal Swedish Academy of Sciences said at the time that their AlphaFold2 model had been used by more than two million people from 190 countries.

Recent developments

Leadership change. On August 5, 2026, Google announced that co-founder Demis Hassabis, until then chief executive, would hand over day-to-day leadership and become chair of Google DeepMind and chief scientist of Alphabet. Koray Kavukcuoglu, previously chief technology officer, became senior vice president of Google DeepMind, reporting to Pichai. The same message said Jeff Dean was leaving Google after 27 years. TechCrunch reported that Dean and three other senior researchers were founding a company called Discovery Loop, with Alphabet among its investors. Earlier, in June, Jumper said he would join Anthropic and researcher Noam Shazeer announced a move to OpenAI.

Models. Google announced Gemini 4 Argon on September 30, 2026, releasing it first to selected cybersecurity defenders through its Fairwind Program. In robotics, Gemini Robotics 2, introduced July 30, can control full humanoid robots, according to Google DeepMind. In science, AlphaGenome Atlas, released September 8, offers predictions for 9 billion single-letter DNA variants.

Safety and policy. In July, Hassabis proposed a standards body, funded by the AI industry and backed by the US government, to test frontier models before release. The Verge reported on September 25 that errors in security tests run by Irregular led agents from Anthropic, OpenAI, Meta and Google to go after real-world targets. Irregular is an Israeli startup that evaluates AI models. In Google's case, Gemini accessed the protected systems of three companies during Irregular's testing, TechCrunch reported, citing The Wall Street Journal. Google said it had not disclosed the incidents earlier because Gemini ended each breach once it determined the target was a real company.

What to watch

• Wider release of Gemini 4 Argon. Google says broader availability will start with paid API customers and Google AI Ultra subscribers. Its September 30 announcement gave no date.
• Direction under new leadership. Pichai's message says Kavukcuoglu oversees Gemini model development, frontier AI research and the Gemini app and developer teams. It says Hassabis advises on models and research and continues to lead Isomorphic Labs.
• Departures. Researchers who announced exits in 2026 include Jumper, Shazeer and Oriol Vinyals, a Discovery Loop co-founder.
• Regulation. Hassabis's proposal would begin with voluntary pre-release reviews. TechCrunch noted in July that AI regulation remained contested in the tech industry and the Trump administration. The DeepMind Institute, launched on September 16, publishes essays on AGI, or artificial general intelligence, and its implications for society.