AI Designs Active Viruses in Lab, Raising New Ethical and Safety Questions

A recent study by researchers from Stanford University and the Arc Institute demonstrated that artificial intelligence (AI) can design genetic codes for bacteriophages, viruses that infect bacteria, and successfully create active viruses in the laboratory. Published last week in the journal Science, the study showed that AI models generated hundreds of new bacteriophage designs, with 16 out of 285 tested designs producing functional viruses capable of infecting and replicating within E. coli bacteria. Some of these newly designed phages were effective against bacteria resistant to the original phage.

Professor Ran Nir-Paz, an infectious disease expert at Hadassah Medical Center and clinical leader of Israel's Center for Phage Therapy, explained that the AI models, named Evo 1 and Evo 2, were trained on vast genetic data to learn the language of DNA and RNA, enabling them to generate new genetic sequences analogous to how language models produce text. This approach builds on earlier synthetic biology breakthroughs, such as the creation of minimal bacterial genomes designed to self-replicate.

While the technology holds promise for developing new treatments against antibiotic-resistant bacterial infections, experts caution that the field is still in its infancy. Only a small fraction of AI-generated designs currently result in viable viruses, and much more data and refinement are needed before reliable applications can be realized. Moreover, the potential for misuse or accidental creation of harmful biological agents raises significant safety and ethical concerns.

Professor Tal Brosh, head of infectious diseases at Assuta Ashdod Hospital, emphasized that bacteriophages naturally infect bacteria and pose no direct threat to humans. However, he warned that synthetic biology combined with AI could be exploited to engineer more virulent or vaccine-resistant pathogens, either intentionally or through laboratory accidents. He cited the example of synthetic recreation of the horsepox virus as a precedent for such risks.

The debate extends to regulatory challenges, balancing scientific freedom with the need to prevent dangerous applications. Current AI models include safeguards against misuse, but these can be circumvented through indirect queries. Despite these risks, AI-driven synthetic biology also offers opportunities to accelerate vaccine development, improve drug discovery, and deepen understanding of biological functions that remain poorly understood. The study highlights the narrowing gap between computer-generated genetic code and living biological systems, underscoring the growing responsibility to manage this powerful technology safely.