AI creates 16 new viruses that never existed in nature after learning DNA’s pattern from 9 trillion nucleotides — experts warn such applications are way ahead of necessary guardrails
An AI model has just created new viruses that never existed in nature. According to a New York Times report, researchers trained a genomic AI model to design complete DNA sequences for viruses, then chemically built the results and watched some come to life. Of 285 AI-generated viral genomes tested, 16 successfully assembled into functioning viruses capable of infecting bacteria and reproducing.
In the study, published in the journal Science, researchers at Stanford University and the Arc Institute trained genomic AI models called Evo on trillions of nucleotides — the individual building blocks of DNA found in every living organism — allowing it to learn statistical patterns in how biological DNA is arranged. Much as a language model learns which combinations of words tend to make sense, Evo learned which combinations of nucleotides tend to produce biologically meaningful sequences.
Scientists have been synthesizing viruses for decades, typically to study how they work and to test antiviral drugs and vaccines. In these projects, researchers typically use the genetic sequence of a virus that already exists to manufacture new ones. Basically, they use existing genomes — genetic sequence data that holds information on how proteins come together to form an organism — replicating and editing this data.
Every organism comprises nuclotides/DNA. The genome provides the instructions for exactly how these building blocks form to create that specific organism. We can copy the data and replicate the organism, aka cloning. However, because trillions of possible genomes exist, it has been statistically impossible for humans to study enough of them to know the formula for how nature puts these building blocks together to form an organism. The blocks were known, but the formula for arranging them in a way that created viable genes wasn’t. Until now.
After training the AI model on over 9 trillion nucleotides spanning over 128,000 genetic sequences drawn from millions of animals, plants, microbes, and viruses, it was able to discover patterns and use those patterns to design new genes that could instruct cells to make specific proteins. Using the pattern, the scientists wanted to see if the AI could create a blueprint for a new, simple organism, such as a virus that contains only a few thousand building blocks. For context, humans contain over three billion.
The researchers then trained the Evo AI model on the 11 genes and 5,386 nucleotides of the Phi X-174 virus and about 15,000 of its closest relatives. The virus was chosen because it has been studied for over a hundred years and is known to attack only E. coli bacteria. After studying the virus’s genome, the model proposed 700,000 new potential genomes, which the scientists trimmed down to 285. The scientists synthesized potential organisms using these instructions, out of which 16 resulted in new viable viruses that have never existed in nature. Like their relative, Phi X-174, the viruses were observed attacking E. coli bacteria and reproducing. It’s important to note that the new viruses are quite similar to the source. The building blocks are very much the same, with some mutations in how they are arranged.
The potential applications of the study can be far-reaching. The scientists demonstrated one of them via a resistance experiment. They first generated three E. coli strains that had evolved resistance to natural Phi X-174. The researchers then exposed those resistant bacteria to cocktails made from the AI-designed phages. Those phage populations evolved during the experiment and ultimately overcame resistance in all three bacterial strains.
The scientists say the new viruses were completely harmless to humans, noting that replicating the same results in pathogens known to affect humans would be a different ballgame. They were also careful not to train the AI on any data from organisms known to affect humans. However, the study inadvertently shows that highly dangerous applications are a possibility. Experts worry that such studies are way ahead of necessary guardrails and regulations. AI has also been known to fly off the rails autonomously. An OpenAI agent recently went rogue and hacked Hugging Face.
