In brief
Using AI models Evo1 and Evo2, researchers at Stanford designed and synthesised 16 novel viruses that successfully infected and killed E. coli in petri dishes. The work marks the first time AI has built fully functional viral genomes, offering a new route to phage therapy against drug‑resistant bacteria. However, experts warn that the same technology could be used to create harmful pathogens, prompting calls for tighter biosafety safeguards. The study illustrates both the transformative potential of synthetic biology and the need for responsible stewardship of powerful generative tools.
The first time a generative AI has built a complete viral genome, researchers at Stanford have shown that software can design bacteria‑killing viruses that thrive inside a petri dish. 16 novel phages were synthesized from 302 AI‑generated candidates, each one proven to replicate and kill a strain of E. coli. The endeavour, published in the journal Science, opens a gate to new drugs for infections that skirt the limits of antibiotics and pushes forward the dream of computer‑designed biology.
Evo1 and Evo2 work like language models, but instead of predicting words they predict what makes a biological ‘sentence’ work. Trained on thousands of genetic sequences from viruses, bacteria, plants and people, the AI drills down to the right pattern of DNA that tells a phage where to head and how to hijack a bacterial cell. The study used phages – viruses that prey only on bacteria – so no risk of harming people.
When the first 16 phages burst onto a slice of bacterial lawn, the team watched in real time the tiny clouds of infection grow into clear spots of death. “We started to see clear spots and it was simply amazing,” says PhD student Samuel King. The result sparked wild applause in the lab. Scientists now hope that such tailor‑made phages could serve as rapid treatment tools against the ever‑growing plague of antibiotic‑resistant infections.
But the breakthrough is not without warning. Dr Thomas Inglesby and Dr Moritz Hanke of the Johns Hopkins Center for Health Security emphasised that the technology can cross a threshold of “urgent biosafety and biosecurity questions.” They argued that the same models that can create harmless phages could be coaxed into designing viruses capable of infecting complex organisms, a possibility that is “no longer a question of existence but of use.” To mitigate risk, the Stanford team deliberately excluded any genome capable of infecting eukaryotes and confined the work to a high‑security laboratory.
The prospects of leveraging AI to write genomes extend beyond viruses. Professor Marc Güell from Spain dubbed the study a “turning point” that turns “genome language models” into a way of learning evolution’s hidden design rules. As AI‑generated biology becomes more sophisticated, the field may soon tackle larger organisms, with implications for enzyme engineering, gene therapy and new materials. Yet each step further into synthetic life comes with a parallel step in responsibility, according to biobiosecurity experts who urge strict oversight and open science collaboration.
For now, the synthetic phages stand as a proof of concept: a clear sign that computers can design living entities, and a reminder that the same algorithms can be turned against life. The next chapter will be written in the laboratory, on the internet, and ultimately, in policy rooms: how to wield a new power that can heal or harm with equal swiftness.
