Research

AI-generated text

Stanford team uses genome language models to design functional bacteriophages, raising governance concerns

Researchers at Stanford applied genome language models—AI tools analogous to chatbot LLMs—to design bacteriophages and produced the first functional viruses created by AI.

Stanford team uses genome language models to design functional bacteriophages, raising governance concerns

Researchers at Stanford University used genome language models—AI tools analogous to large language models used in chatbots but trained on genetic sequences—to design viruses. The team produced bacteriophages, viruses that infect bacteria and are already used in some contexts to treat bacterial infections. In laboratory tests, a cocktail composed from the AI-designed phages killed Escherichia coli strains that were resistant to natural phages.

Experimental details and outcomes

The researchers synthesized nearly 300 designed genomes; of those, only 16 produced functional phages. To reduce risk, the training data explicitly excluded human and animal viruses. The results were published in Science, and the paper includes an explicit recommendation to involve safety experts in further work.

Why this matters

This work demonstrates that generative AI can now compose viral genomes—effectively creating the genetic "source code" of self-replicating biological entities. Although the particular phages created in this study are not intended to infect humans or animals, the technical capability to design functioning viral genomes has been shown. That creates a precedent: capability has arrived faster than governance.

Governance and expert reactions

Scientists at Johns Hopkins noted bluntly that the ability to compose viral genomes with generative AI exists today, while the governance frameworks to manage that capability do not. Critics and commentators emphasize that developing policy, oversight, and safety practices has not kept pace with the technical advance, and that such measures are now urgently needed to determine who may run these tools and under what conditions.

Conclusion

The Stanford study is a notable proof of concept: genome language models can produce bacteriophages that work in the lab, offering potential therapeutic avenues against resistant bacterial strains. At the same time, it highlights a gap between emerging biological design capabilities and the regulatory and ethical structures required to ensure their safe, responsible use.