Google—the same company that tracks your searches, censors your results, and hoards your data—now claims to have mapped every possible mutation in human DNA. And it's keeping the price tag for who gets to profit off your genome a secret.
AlphaGenome Atlas, released Tuesday by Google DeepMind, is a precomputed catalogue predicting what each of the 9 billion possible single-letter DNA substitutions is likely to do to the machinery that switches genes on and off. The database weighs in at roughly 1 petabyte—30 times larger than DeepMind's celebrated AlphaFold database. The promise: faster research into genetic diseases. The catch: Google decides who gets access, and on what terms.
Pushmeet Kohli, DeepMind's vice president for research, told reporters this marks "the first time any researcher in the world can access a comprehensive map of the human genome and its variations by simply opening a browser." Noncommercial researchers can access Atlas starting now. Commercial users—drug companies, biotech firms—will have to license it through Google Cloud. When? "Soon." For how much? DeepMind wouldn't say. Kohli declined to specify commercial licensing terms, according to Fortune.
Follow the money. DeepMind's sister company, Isomorphic Labs, which uses AI for drug discovery, will also need a commercial license, Kohli said—though one might reasonably ask whether the firewall between Alphabet subsidiaries is as solid as the press briefing suggests. Demis Hassabis, DeepMind's cofounder, is stepping away from running the AI lab to lead Isomorphic Labs full time. The man who won a Nobel Prize for AlphaFold is now positioned to monetize the genome.
Notably, Scientific American buried the most important caveat: AlphaGenome Atlas is "far less accurate" than AlphaFold's database. It's meant as a "starting point," says Žiga Avsec, DeepMind's genomics lead—not a final answer. That didn't stop The Verge from framing the release as a tool that could "transform our understanding of biology" and "pave the way for new treatments." Fortune leaned into the therapeutic promise too, calling it the unfinished business of the Human Genome Project. None of the three outlets raised a single question about who defines what counts as a genetic "defect," what oversight exists, or what it means when a surveillance capitalist effectively claims stewardship over the map of human variation.
The atlas assigns each variant an "AlphaGenome Variant Impact" score—a single metric ranking which mutations supposedly matter most. Each variant comes with roughly 27,000 individual predictions about effects on gene expression, protein production, and more, across hundreds of cell and tissue types. That's an enormous amount of interpretive power concentrated in one company's model, trained on public data, now monetized through private licensing.
AlphaFold's database was free. This one isn't. The shift tells you everything about where Google's AI-for-science program is heading: public data in, paywall out.
The question nobody in Washington is asking is simple: when a single company owns the predictive map of every possible human genetic variant—and the scoring system that ranks them—who decides what's a defect worth treating, and who gets left on the cutting room floor?
