Image Source: The New York Times
Scratch the surface of a major scientific breakthrough and an unsettling question emerges: scientists have created the first functioning viruses designed by artificial intelligence. The achievement could open new routes to medicines for persistent bacterial infections, but it has also triggered urgent debate about biosafety, laboratory controls and the future of generative AI.
The viruses are bacteriophages, or “phages”—specialized viruses that infect bacteria rather than people. Researchers around the world already use phage therapy experimentally and, in some cases, clinically, particularly when patients have infections that no longer respond to antibiotics or naturally occurring phages.
Scratch the Surface of AI-Designed Viruses
Dr Brian Hie, a chemical engineer at Stanford University in California, led the work with colleagues. Their research used genome language models called Evo1 and Evo2. Similar in concept to large language models that generate text, these systems were trained to analyze and produce genetic sequences.
The models studied genetic information from about 2 million bacteriophages. To reduce the risk of creating dangerous organisms, the researchers deliberately excluded genetic data from viruses known to infect humans, animals or plants.
The AI produced thousands of potential viral genomes. Scientists selected nearly 300 candidates for laboratory testing, placing their genetic code into bacteria. The bacteria then read the sequences and produced newly designed bacteriophages.
The process was far from perfectly efficient. Only 16 of the candidates proved viable. However, the successful viruses showed an important advantage: combined in a cocktail, they rapidly killed two strains of E coli that had resisted natural bacteriophages in laboratory tests.
Powerful Medical Promise, Critical Safety Questions
The researchers said the ability to rapidly design phages and tailor them to specific bacteria could transform phage therapy. It may eventually help doctors respond more quickly to persistent infections and develop treatments that overcome resistance.
In their research, published in the journal Science, the scientists described the work as a way to expand biotechnology tools. The approach could be especially valuable as antibiotic resistance continues to make certain bacterial infections increasingly difficult to treat.
Yet the same capability raises serious concerns. In an accompanying article, Professor Tom Inglesby and Dr Moritz Hanke of the Center for Health Security at Johns Hopkins University warned that generative AI can now compose functioning viral genomes, while the systems needed to govern that ability safely remain underdeveloped.
“The ability to compose viral genomes using generative AI now exists; the governance to safely steer it does not,” they wrote.
Inglesby and Hanke said the experiment demonstrated a significant technical milestone even though bacteriophage genomes are relatively small. They cautioned that researchers should not extend the same approach to pathogens capable of infecting humans, animals or plants. Such organisms could potentially evade existing countermeasures, they warned.
Experts Call for Layered Biosecurity Controls
Tom Ellis, a professor of synthetic genome engineering at Imperial College London, described the research as impressive but stressed that bacteriophages are among the simplest genomes to construct. Designing a complex human pathogen would be considerably more difficult, and the study does not show that such a step is currently practical.
Ellis also argued that the immediate threat may be overstated. Existing pathogens modified through gain-of-function research could pose a more realistic danger than entirely new organisms designed from scratch. Still, he said better controls over genetic data and restrictions on manufacturing sequences that appear dangerous would be important.
Dr Filippa Lentzos, a reader in science and international security at King’s College London, urged policymakers to look beyond the AI model itself. She said the most important intervention point may be DNA manufacturing, where screening systems can identify suspicious genetic sequences before they are produced.
- Stronger safeguards around AI model development and access
- Responsible research reviews before experiments begin
- Screening of DNA sequences ordered from manufacturers
- Established laboratory biosafety and biosecurity procedures
The debate is therefore not simply about whether AI can design viruses. It is about ensuring that medical innovation develops alongside effective oversight. As researchers scratch deeper into synthetic biology, the potential benefits are substantial—but so is the responsibility to prevent misuse.
Frequently Asked Questions
What are bacteriophages?
Bacteriophages are viruses that infect bacteria. Because they generally target bacterial cells rather than human cells, they are being studied as tools for treating infections, especially those resistant to antibiotics.
What did the AI models create?
The Evo1 and Evo2 genome language models generated genetic sequences for bacteriophages. Researchers tested nearly 300 candidates and identified 16 viable viruses.
Could AI-designed viruses treat infections?
The laboratory results suggest they could eventually support personalized phage therapy. A cocktail of the new viruses killed resistant E coli strains in testing, but further research and clinical evaluation are required.
Why are scientists concerned?
Experts worry that similar methods could eventually be applied to more dangerous viruses. They are calling for safeguards covering AI access, research approval, DNA synthesis screening and laboratory security.
