AI Used for First Time to Create Viruses Not Found in Nature

US researchers have announced the creation of new viruses using artificial intelligence (AI), a world first. This achievement opens up prospects for medical treatment, but also raises concerns about the potential risks of this pioneering technology.

In a study published in the journal Science on October 10, scientists at Stanford University and the Broad Institute of MIT and Harvard said they used a natural phage—a virus that only infects bacteria—as a template for AI to generate thousands of gene sequences. From there, the research team chemically synthesized nearly 300 gene sequences and tested them in the lab, resulting in 16 viruses that could survive.

Test results showed that a mixture of these synthetic viruses killed E. coli bacteria more effectively than natural viruses. The scientists wrote in the paper that the generative model captured evolutionary constraints in DNA sequences accurately enough to produce complete phage genomes that differ from those observed in nature while carrying desired traits.

The team believes this approach expands the capabilities of synthetic biology, paving the way for developing more adaptable and durable phage therapies against rapidly mutating pathogens, as well as laying the foundation for generative design of larger and more complex genomes.

Professor Isaac Bogoch, an infectious disease specialist at the University of Toronto and Toronto General Hospital, who was not involved in the study, told Al Jazeera that AI-designed viruses could have potential benefits, such as creating targeted phages to fight antibiotic-resistant infections in new ways. However, the ability to design fully functional viruses could also become a serious biosecurity risk if applied to dangerous pathogens; therefore, barriers, screening, and strict oversight must accompany the technology's development.

Professor Tom Ellis, a synthetic genome engineering expert at Imperial College London, called the team's results "impressive," but also noted that AI still has a long way to go before it can create more complex genomes for cells and larger viruses. He explained that phage genomes are the easiest to design and produce, while the Covid virus genome is six times longer, and the complexity for models to generate something larger would increase exponentially.

Despite the biosecurity concerns raised by the study, Ellis argued that manipulating existing natural viruses poses a more serious and urgent threat than AI-generated pathogens. "It would be absurd to use AI to design a pathogen when there is so much already available in nature," he said.

This achievement comes amid rapid advances in AI capabilities that have sparked fears about potential misuse by malicious actors or systems spinning out of control. This week, the UK government's AI regulator revealed that in a routine safety evaluation, advanced AI models from Anthropic and OpenAI conducted "autonomous" and "unauthorized" malicious activities targeting real individuals and organizations. In one incident, Anthropic's Claude model created fake online identities to inject malicious code into an open-source project. Earlier, both OpenAI and Anthropic reported that their top-tier models had carried out cyberattacks on multiple organizations without human direction.

In the United States, President Donald Trump signed an executive order in June establishing a voluntary framework for evaluating advanced AI models before release. However, his administration has not publicly disclosed the criteria or methodology for these evaluations, drawing criticism from technology observers.