AlphaGenome gives researchers a new way to study the effects of billions of possible DNA variants.
Co-Scientist uses Gemini-based AI agents to create, test and refine scientific hypotheses.
Hassabis now has a broader Alphabet science role while Isomorphic Labs expands its AI drug discovery work.
Demis Hassabis has moved into a new role at Alphabet, while continuing to focus on using advanced AI to address difficult scientific problems. On August 5, 2026, Hassabis became Chair of Google DeepMind and Chief Scientist of Alphabet. Koray Kavukcuoglu took responsibility for day-to-day Google DeepMind operations and Gemini development. Hassabis also remains CEO of Isomorphic Labs, where AI has a direct role in drug discovery.
AlphaFold remains the clearest example of Hassabis’s science-first approach. Google says the AlphaFold database has reached more than 3 million researchers in more than 190 countries. The database now contains more than 200 million predicted protein structures. AlphaFold-related research has produced more than 35,000 citations, while more than 200,000 papers have used AlphaFold 2.
AlphaFold 2 addressed a protein structure problem that had challenged science for about five decades. Hassabis and John Jumper received the 2024 Nobel Prize in Chemistry for work tied to AlphaFold 2. That success raised a larger question: Can AI help science move from prediction to discovery?
AlphaGenome takes the same basic idea into genetics. On September 8, 2026, Google DeepMind announced AlphaGenome Atlas, a resource with predictions for all roughly 9 billion possible single-letter DNA variants in the human genome. The dataset reaches about 1 petabyte, more than 30 times the size of the AlphaFold database.
AlphaGenome helps narrow a huge search space. Its AlphaGenome Variant Impact score gives researchers a way to rank variants by possible biological effect. A University of Exeter study used AlphaGenome Atlas with data from more than 54,000 UK Biobank participants. One analysis cut a region with 526 candidate variants down to four variants that offered a clearer path for biological study.
Also Read - Meta Muse AI: 15 Useful Prompts & Use Cases You Should Try
Google DeepMind introduced Co-Scientist in May 2026. The system uses Gemini and several AI agents to create, test, critique and refine scientific hypotheses. DeepMind says more than 100 institutions have worked with the system.
Early research has covered antimicrobial resistance, plant immunity, liver fibrosis, ALS, infectious disease and molecular biology. In one liver fibrosis case, DeepMind says Co-Scientist found drug-reuse candidates that researchers had overlooked. One candidate blocked 91% of a fibrosis-associated response in laboratory tests.
Such results still need careful validation. A strong AI hypothesis does not equal a proven treatment. A safe medicine still requires lab work, clinical trials and regulatory review.
Google now wants Gemini to serve as a broader scientific workbench. Gemini for Science combines Co-Scientist with systems such as AlphaGenome, AlphaFold and AlphaEvolve. It also connects researchers with more than 30 life-science databases and tools.
Instead of one model for one scientific task, Google aims for a general AI system that can use specialized models, databases and research tools. WeatherNext 3, launched on September 3, 2026, extends the same idea to weather prediction. Google says the system delivers precipitation forecasts that are 50% more accurate one day or more ahead than those from its earlier system.
In December 2025, Google DeepMind announced plans for its first automated science lab in the UK, with an initial focus on materials science. The lab aims to combine Gemini with robots that can make and test hundreds of materials per day.
That model creates a tighter research cycle. AI can propose an idea, a robot can test it, instruments can measure the result, and the data can guide the next hypothesis.
Also Read - CRO 2027 Checklist: 10 Revenue Priorities for AI-Driven Growth
Hassabis’s work at Isomorphic Labs adds another major part to the strategy. In May 2026, Isomorphic Labs raised USD 2.1 billion in Series B capital. The company plans to scale its AI drug-design platform and move its therapeutic pipeline toward clinical development. Its partners include Novartis, Eli Lilly and Johnson & Johnson.
The larger vision now looks less like a single breakthrough model and more like a scientific engine. AlphaFold can help map proteins. AlphaGenome can help study DNA variants. Co-Scientist can help create hypotheses. Gemini can connect those systems with research tools, while automated labs can test ideas in the physical world.
Hassabis’s new position at Alphabet gives that vision a wider reach. His work now sits at the intersection of frontier AI, scientific research and medicine. Better AI can help scientists search spaces that human research alone cannot explore at the same scale.
1. Who is Demis Hassabis?
Demis Hassabis is a co-founder of Google DeepMind and a major figure in AI research and scientific discovery.
2. What is AlphaFold?
AlphaFold is an AI system that predicts protein structures and has become a major tool for biological research.
3. What does AlphaGenome do?
AlphaGenome predicts the effects of DNA changes and helps researchers identify genetic variants that may have important biological effects.
4. What is Co-Scientist?
Co-Scientist is a Gemini-based system that helps create, test, critique and refine scientific hypotheses across several research fields.
5. How does Isomorphic Labs fit into Hassabis’s work?
Isomorphic Labs applies AI to drug discovery and aims to use advanced computational tools to help develop new medicines.