New AI-generated viruses spark biosafety fears

New technology to create synthetic viruses that do not exist in nature with artificial intelligence is scaring biosafety experts, who worry that the process could spark the next global pandemic. Manipulating the genetic structure of naturally occurring viruses is not new, but using AI to create entirely new viruses poses new safety concerns and puts scientists and policymakers in uncharted territory. Scientists at Stanford University published a new paper on Thursday in the journal Science in which they effectively taught AI to recognize DNA patterns and, from there, effectively grow viruses never before seen in nature. The viruses were then able to infect bacterial cells, demonstrating they were viable. The new use of AI has sent shockwaves through the life science research community and given rise to concern that misuse of the new technology, unintentional or otherwise, could cause a public health crisis similar to the scale of the COVID-19 pandemic. Richard Ebright, a molecular biologist at Rutgers University, told the Washington Examiner that AI has the capacity to make the technical process of genetic engineering of viruses “faster, cheaper, and easier.” Ebright has been an outspoken critic of gain-of-function research, which is the genetic manipulation of viruses to make them either more infectious or more deadly. He has also been one of the leading advocates of the theory that SARS-CoV-2, the virus that causes COVID-19, was created in a research laboratory in Wuhan, China, the epicenter of the initial pandemic outbreak. When asked about the Stanford experiment, Ebright said the new capacity of AI only strengthened his desire to see governments around the world enact “strict oversight or an absolute ban on dangerous gain-of-function research.”“With AI rapidly accelerating dangerous gain-of-function research, the next lab-generated pandemic almost surely will emerge within the next decade,” Ebright said.The Stanford researchers did take precautionary measures in their creation of the AI model, named Evo, with the intention of thwarting the technology’s ability to create viruses that infect humans. Evo is similar to other large language models, or LLMs, such as ChatGPT or Claude. Much like spoken or written languages, DNA consists of a string of nucleotides that, when strung together following a basic set of rules, encodes the instructions for building proteins and other molecules.Just like other LLMs, Evo is able to produce content based only on the material it is given access to for studying. The Stanford team did not give Evo data about viruses that infect humans or other animals, plants, and fungi to prevent the AI model from running wild.With those precautions in place, the Stanford researchers see Evo as being a game changer for some of the largest problems in infectious disease medicine, like antibiotic resistance.Samuel King, one of the members of the Stanford team, told Financial Times that he sees the experiment as proof of concept for phage therapy, or using viruses to kill bacteria, that could be vital in a world where antibiotics are becoming less effective in killing dangerous infections.“With more evidence coming out on the emergence of scary antibiotic-resistant pathogens, we will need innovative alternative solutions — and phage therapy is certainly one,” King said.But no new technology comes without risks.Dr. Mortiz Hanke and his colleague, Dr. Thomas Inglesby, both from the Johns Hopkins Center for Health Security, coauthored an opinion piece in the same journal that published the Stanford study, cautioning the scientific community on the use of the technology without proper guardrails.HOW DID FAUCI’S DIARIES MAKE IT PUBLIC? NO ONE WILL SAYEffectively, using AI lowers the bar for entry for virus creation, but policymakers are lagging behind the rapid scientific developments necessary to prevent the emergence of the next deadly superbug.“Although this is promising for life sciences applications, it also raises urgent biosafety and biosecurity questions,” Hanke and Inglesby wrote. “The ability to compose viral genomes using generative AI now exists; the governance to safely steer it does not.”