Scientists create first AI-designed viruses, raising safety concerns

Scientists have created the first viruses whose genomes were designed by artificial intelligence (AI), a milestone that could boost the development of new medicines but also raises questions about keeping the technology safe.

The viruses are bacteriophages — pathogens that only infect bacteria and are already used worldwide to treat patients with persistent infections. In laboratory tests, a cocktail of the AI-designed phages killed E. coli strains that were resistant to natural bacteriophages.

Dr Brian Hie, a chemical engineer at Stanford University in California, and colleagues used genome language models — the genetic equivalent of large language models behind AI chatbots — to design functioning genomes. The models, named Evo1 and Evo2, were trained on genetic data from two million bacteriophages. To reduce the risk of creating dangerous viruses, the genetic code of viruses that can infect plants, humans or other animals was deliberately excluded from the training data.

The AI generated thousands of potential genomes, of which researchers selected nearly 300 to build in the lab. These were inserted into bacteria, which read the genetic code and produced the new phages. The process was inefficient: only 16 phages proved viable. However, a cocktail of them quickly overcame resistance in two distinct E. coli strains.

The team wrote in the journal Science that the ability to rapidly design genomes and tune them for specific bugs while overcoming resistance could transform phage therapy and expand biotechnological toolkits. They also noted important biosafety, biocontainment and biosecurity considerations, urging others who design whole genomes to consult safety and security professionals throughout their projects.

In an accompanying article, Prof Tom Inglesby and Dr Moritz Hanke from the Center for Health Security at Johns Hopkins University reinforced the warning. They wrote that while AI-driven genome composition holds promise, current governance is not adequate to steer it safely. They cautioned that pursuing work on pathogens that could infect humans, animals or plants should not be done, as such genomes might encode new pathogens that existing countermeasures cannot contain.

Tom Ellis, a professor of synthetic genome engineering at Imperial College London, described the work as impressive but noted how difficult it would be to produce more complex genomes. He said the threat from full AI design of a viral or bacterial genome is overblown, because modifying existing pathogens is far easier and poses a more realistic pathogenic threat.

Dr Filippa Lentzos, a reader in science and international security at King's College London, said the most important intervention point is DNA manufacturing. She argued for a layered approach to governance, including safeguards around model development and access, responsible research review, synthesis screening, and established laboratory biosafety and biosecurity measures.