AI-designed viruses: what the first successful lab test means for biosecurity

It sounds like science fiction: an artificial intelligence model designing a virus from scratch that actually works in a laboratory. But according to the BBC, that is exactly what researchers have achieved — 16 viruses whose genetic code was designed entirely by an AI system successfully infected their targets in lab tests.

The viruses in question are bacteriophages, or phages for short — tiny biological entities that infect and hijack specific bacteria, not humans. Phages have existed in nature for billions of years. Researchers trained an AI model on thousands of existing phage genomes, then asked it to generate entirely new designs that don't exist in nature.

The approach echoes tools like AlphaFold, which predicts protein shapes, but applied to full genomes instead. The model learned the patterns that billions of years of evolution left in phage DNA, then recombined those patterns to propose brand-new genetic sequences that might function as working viruses.

Scientists synthesized dozens of these AI-generated designs and tested them in the lab. According to the BBC, 16 of them actually worked — meaning they successfully infected their target bacteria and replicated. For a biological system designed entirely by software, researchers describe that success rate as a notable milestone.

The potential benefits are significant. Antibiotic-resistant bacteria are a growing global public health threat, and phage therapy — using targeted viruses to kill resistant bacteria — is one proposed solution. An AI that can rapidly design effective phages could substantially speed up the development of new treatments for infections that no longer respond to conventional antibiotics.

But the same capability raises serious biosecurity questions. If a model can design a working bacteriophage, similar methods could in theory be misused to help design more dangerous pathogens. Researchers stress that the phages used in this study cannot infect humans, but experts note that clear rules for overseeing this kind of dual-use research are still lacking.

Reaction within the scientific community has been mixed. Some researchers view the work as a landmark moment for synthetic biology, while others argue that access to these genome-design tools, and how such research is published, needs far closer scrutiny. According to the BBC, the team behind the study says safety review was built into the design process from the outset.

The development fits a broader trend of AI becoming embedded in biology. From protein-folding predictors to systems that propose new drug candidates, AI tools have become a standard part of laboratory science over the past several years. Genome-level design is seen as the next stage of that trend.

For now, scientists are proceeding with cautious optimism. They plan to keep developing phages as medical tools while acknowledging that regulatory frameworks need to evolve in parallel with these increasingly powerful design capabilities.

The episode is likely to be a preview of a wider conversation: as AI systems get better at designing biology rather than just analyzing it, the question of who gets to build what — and who checks it first — is only going to grow more urgent.