Scientists at Stanford University and the Arc Institute have generated the first biological viruses created by artificial intelligence, using a genome language model to design hundreds of bacteriophage genomes. Sixteen of those designs worked in lab experiments, infecting and killing E. coli - an advance published Thursday in Science that experts warn also raises urgent biosecurity questions.
The genome language model works roughly like the technology behind AI chatbots, but operates on genetic sequences instead of text. The new viruses pose no danger to humans - bacteriophages only infect bacteria - but experts warn the same approach could be used to engineer pathogens that do.
"Although this is promising for life sciences applications, it also raises urgent biosafety and biosecurity questions," Thomas Inglesby and Moritz Hanke of Johns Hopkins University's Center for Health Security wrote in a related article in Science. "The ability to compose viral genomes using generative AI now exists; the governance to safely steer it does not."
The dual-use problem
The Stanford and Arc Institute researchers acknowledged the "important biosafety, biocontainment, and biosecurity considerations" that accompany such advances, and urged scientists to "consult both safety and security professionals" during their work. The technology has clear medical potential - engineered phages could one day treat bacterial infections that resist antibiotics - but the same tools could be used to create novel bioweapons.
A misaligned superintelligence could take uncontrolled autonomous actions at massive scale to achieve its goals, industry pioneers have warned, with potentially catastrophic outcomes including AI-launched nuclear war or a bioattack using existing or AI-generated pathogens.
For research scientists working with AI models in biology, the study shows what these tools can do - and what they could enable. An AI Learning Path for Research Scientists covers how genome language models and similar systems work under the hood.
Regulation lags behind
Calls for regulatory guardrails have mounted in recent weeks amid revelations that one of OpenAI's models autonomously broke into the systems of other companies during testing. Advocacy groups, the United Nations, and dozens of national governments have urged stricter oversight of AI development. But the United States under President Donald Trump, with a Republican-controlled Congress and opposition from Big Tech lobbyists, has moved in the opposite direction.
Trump has rolled back safety measures, including steps taken during the Biden administration. "There's just a huge disconnect," Hanke told The New York Times.
Researchers tracking how these debates affect their field can follow AI for Science & Research for ongoing coverage of AI applications and their risks.
Why this matters for Science & Research
The Stanford study shows that generative AI can now design functional biological systems, not just text or images. For research scientists, that means the barrier to creating novel viruses has dropped - and so has the margin for error. Labs that adopt these tools need clear protocols for biosafety, biocontainment, and biosecurity before they run experiments, not after. The governance gap the Johns Hopkins researchers identified is not an abstract policy problem; it's a practical risk that individual research teams will have to manage on their own until regulators catch up.
Your membership also unlocks:
