An artificial intelligence created brand–new living viruses 

What if machines could invent infectious viruses nature never made? U.S. researchers pulled off this AI feat, while specialists fear the same tech might forge deadly human pathogens with no guardrails in place.

For the first time in scientific history, researchers have successfully used artificial intelligence to create completely new viruses that do not exist in nature. Released in the journal Science, the work marks a key shift from copying existing viral DNA. “This is an important milestone,” said Patrick Cai, a synthetic biologist at the University of Manchester. Rather than reconstructing known pathogens, the research team taught an AI system to identify structural patterns inside natural DNA, then compose original genetic blueprints to build brandnew viral life forms.

Nevertheless, the innovation also brought security problems. The same technology could one day be used to create dangerous new bioweapons. Even as bioscience advances at a rapid pace, national governments and global scientific communities have struggled to keep up. They have been slow to develop guardrails and clear policies that prevent the design of deadly viruses.

The core of the project is the AI model Evo. Scientists fed it massive genetic datasets collected from millions of animals, plants, microbes and viruses. This works because DNA is strikingly similar to a book and has its own rules of grammar. A gene includes hundreds of nucleotides, which are arrayed like letters. Sequences that violate this hidden biological grammar make the result useless. By processing trillions of genetic data points, Evo independently learned these complex biological rules.

Using this biological knowledge, Evo generated 700,000 potential viral genomes. Researchers narrowed them down, physically synthesising 285 selected DNA sequences for lab tests. After inserting these artificial genomes into bacterial cells, the team confirmed that 16 of the AIwritten sequences turned into viruses. “They’re not just sickly versions of stuff that already exists,” noted Oliver Crook, protein chemist at the University of Oxford. Several AIbuilt viruses even multiplied faster than their natural counterpart.

Potential risks were never ignored by the researchers when training Evo. The team did not provide the data about infectious viruses, and excluded similar viruses that infect other creatures. As a result, the current version of Evo cannot generate viral genomes that are possible to threaten human health. However, unpredictable future dangers still puzzle experts. “What is the risk of what I’ve never seen before?” asked Dr. Moritz Hanke.