AI-Designed Viruses Could Become a New Weapon Against Drug-Resistant Bacteria

Virologists and computer scientists have used artificial intelligence to design entirely new viruses, but not ones that infect humans or animals. Instead, the AI models created custom bacteriophages, viruses that hunt and destroy bacteria, with the goal of improving future medical treatments.

The project, involving researchers from Stanford University and several partner institutions, shows that AI can move beyond analysing existing biological data. It can now help design new biological entities from scratch, marking a significant shift in how modern biotechnology is conducted.

The researchers focused specifically on bacteriophages that infect strains of E. coli. The work could ultimately help scientists develop new ways of targeting drug-resistant bacteria, while also raising questions about how increasingly powerful biological AI systems should be controlled.

How AI learned to design viruses

The team used specialised genomic AI models known as Evo, trained on DNA sequences rather than words. Much like language models predict the next word in a sentence, these systems learn patterns in genetic code and can propose new DNA sequences that do not exist in nature.

For this experiment, the researchers focused on bacteriophages that infect specific strains of E. coli. By generating and then testing novel phage genomes, they identified viruses capable of attacking bacteria while remaining harmless to humans, animals and plants.

Such AI-generated phages could eventually be tailored to target drug-resistant bacteria. This approach is attracting interest as antibiotic resistance spreads worldwide and traditional drugs fail more often in hospitals and community settings.

Promise for fighting antibiotic resistance

Bacteriophage therapy has been studied for decades as an alternative to antibiotics, especially in Eastern Europe and parts of the United States. However, designing or finding the right phage for a particular bacterial infection has been slow and labour-intensive.

AI could accelerate this process by rapidly exploring vast numbers of possible genetic combinations, highlighting those most likely to infect a chosen bacterial strain. If confirmed by laboratory testing and clinical trials, this could support personalised treatments for infections that currently have few options.

Experts note that AI-assisted design might also help update phage therapies as bacteria evolve resistance. Instead of searching for new natural viruses, scientists could ask models to generate fresh candidates, then refine and test them in controlled conditions.

Biosecurity fears and ethical questions

The same capabilities that make AI powerful for medicine also worry biosecurity specialists. If AI can design beneficial viruses, they warn, future systems could someday help create more dangerous ones, especially as models become more accessible and more capable.

In the Stanford-led work, the team deliberately avoided data from viruses that infect humans and focused on relatively simple bacteriophages. The study did not produce agents that threaten people, but it demonstrated that AI can help invent functional biological entities beyond what nature provides.

Critics fear that as tools for generating genetic sequences become easier to use, advanced biological design might move from high-security laboratories to less regulated settings. That raises questions about who should be allowed to access powerful models and what safeguards are needed.

Regulation and the road ahead

Researchers involved in the project argue that early transparency and strong oversight are crucial. They call for updated rules on publishing genetic designs, stricter controls on DNA synthesis services and clearer global standards for dual-use biological research.

Policymakers and scientific bodies are starting to respond. International organisations and national governments are debating how to classify AI-assisted biological design, and whether existing biosafety and export-control frameworks are sufficient.

For now, AI cannot simply generate dangerous human pathogens on command, and laboratory validation remains a major barrier. Yet the latest experiment signals a turning point, showing that computers can help both decode life and design new forms of it, forcing society to weigh medical benefits against emerging biosecurity risks.