Last updated:
For the first time, AI produced a complete viral genome that works in the lab, killing E. coli, opening up new possibilities in medicine and biotechnology
(Special image)
Artificial intelligence (AI) has already proven that it can generate poetry, essays and images. Now, California researchers say it wrote something much more radical: the complete genetic code of a virus that can infect and kill bacteria. When scientists from Stanford University and the ARC Institute tested hundreds of these AI-designed genomes in the lab, 16 of which produced bacteriophages, viruses, which attack bacteria that they replicated and wiped. E. coli (E. coli), harmless bacteria that are often used as test systems in labs.
The study was published on September 17th on Biorxiv, an online platform where scientists share their research before they are officially reviewed, and its authors have been described as the first generation design of the complete genome.
How did AI write the genome?
This project used an AI system called EVO, designed with the same principles as a large-scale language model. Instead of predicting words in sentences, EVO was trained on around 2 million viral genomes, allowing them to learn about the “grammar” of DNA, the patterns of gene order and composition. The researchers mimicked the model and mimicked the Phix174. It is a small bacteriophage with just 11 genes and about 5,000 DNA letters, and has long been a staple in molecular biology.
From this training, EVO proposed a complete and viable genomic sequence. Some had new or truncated genes, with shuffling gene ordering, not just copying existing phages.
From cords to bacteria killing
Of the thousands of AI proposals, the team led by Brian Hee and Samuel King selected 302 chemically synthesized genomes. We checked whether these DNA strands were introduced into bacterial systems and assembled into functional phages. A key test occurred when some designs produced plaque. E. coli The colonies had been wiped out with phages. In total, 16 designs have been successful from 302, producing viruses that replicated the interior of bacteria and exploded.
According to the team's preprint, some of these phages are infected E. coli Strains that failed to native phix174 indicate that AI produced alternative genetic pathways that could have the same fatal effect.
Why is this outcome important?
For scientists, the major breakthrough is not that another bacteriophage was created, but that AI could generate the overall coherent genome that worked in the system in which it actually lived. Previous efforts in synthetic biology relied on painstaking testing and errors, and one genetic adjustment at a time was manually tested.
Pioneers such as J. Craigventor, who created the first cell using synthetic genomes in 2008, describe the process as a “manual AI version” that sifts through the data and guesses what works. EVO short circuit to perform this processing. MIT Technology Review.
The meaning is broad. Phage has already been studied as an alternative to antibiotics in the treatment of resistant infections, reported Newsweek. The AI-designed version allows you to expand the menu of viruses that can be used for specific bacteria.
The virus also acts as a shuttle for gene therapy, and with a more efficient design, it can improve how genes are delivered to patients' cells. In agriculture, phages are being tested to protect crops from bacterial diseases. By speeding up the design process, AI can turn these niche experiments into more practical tools.
Safety dilemma
Viruses are not considered living organisms, so the results fall short of “life created by AI.” Brian Hee, who led the ARC Institute Lab, called it a step into the generational life of AI, but Samuel King, who led the project, emphasized that even more experimental advances would be needed to reach that point.
Despite these limitations, this study revived the debate about the double use dilemma of biology. The Stanford-ARC team noted that EVOs were trained only with bacteriophages and were tested for non-pathogenicity. E. coli and the PHIX174 have long been considered a safe lab system.
However, outside experts raised more keen concerns. Geneticist J. Craigventer said Newsweek If a similar random virus-enhancing approach was applied to natural PO or charcoal thr, he would have “significant concerns.” Nature Cartin Gebfrich, a biophysicist at Heidelberg University, cited the double use dilemma as not inherent in AI because “you can always use progress for the better or worse direction.”
What's coming next?
As this study is prelint of biorxiv, the next step is formal peer review and independent replication to confirm results. Beyond that, some in the field already imagine how the process can be expanded.
Conclusion
For the first time, AI has produced a complete viral genome that works in the lab. 16 of killed AI designed phages E. coliNatural phix174 containing strains was not possible. Experts such as Jef Boeke, a genomic scientist at Nyu Langone Health in New York, said MIT Technology Review The AI output was impressive, but still far from creating life, it marks an important milestone in biotechnology.
Karishma Jain, editor-in-chief of News18.com, writes and edits opinions on a variety of subjects, including politics and policy, culture, arts, technology and social change in India. Follow her @kar…read more
Karishma Jain, editor-in-chief of News18.com, writes and edits opinions on a variety of subjects, including politics and policy, culture, arts, technology and social change in India. Follow her @kar… read more
September 20th, 2025, 15:40 IST
read more
