Stanford researchers used AI to create 16 viruses not found in nature

Researchers at Stanford University and the Palo Alto-based Arc Institute used an artificial intelligence model to create viruses not found in nature. The first of its kind study, published Thursday in the journal Science, reveals how the Bay Area scientists successfully taught an AI model to create new viruses, which could in turn be a boost for future research on diseases and medicine. 

Read more The best photos from Day 1 of Outside Lands

The viruses made in the lab are not a danger to humans, the researchers insist, but a paper published in the same issue of Science by researchers who were not involved in the study warned that the emerging tech has the potential to be abused, and called for time-sensitive regulations.

“Although this is promising for life sciences applications, it also raises urgent biosafety and biosecurity questions,” the accompanying paper stated. “The ability to compose viral genomes using generative AI now exists; the governance to safely steer it does not.”

Though the process of making a new virus may appear to be fundamentally different from the ways many people are using AI tools today, it’s actually quite similar.

Chatbots like ChatGPT, Claude or Gemini are large language models trained on massive amounts of written content that then predict which combinations of words best answer a question or complete a task. The AI model researchers used to make viruses follows the same process, but with genetic data. 

The Palo Alto researchers’ models are “trained on millions of natural genomes” to produce specific outcomes, according to the study. The AI model is then used to generate viruses modeled after a naturally occurring virus known as Phi X 174, which only infects the E. coli bacteria. Out of the nearly 300 unique viruses created, researchers only deemed 16 new viruses viable after they successfully infected bacteria.

Read more Prices for Uber and Lyft surged after Outside Lands

According to the New York Times, the researchers instituted several safeguards to ensure the model was not capable of creating anything that could harm humans. Researchers did not give the model any information about viruses capable of infecting humans, or viruses that pose a threat to plants and animals. 

Despite the safeguards, the advancement poses the question of whether the technology could be used to create bioweapons such as viruses that present a danger to humans. Though he didn’t downplay the significance of the research, one expert assured the Guardian that likeliness of catastrophe was far off as the researchers’ viruses were biologically simple.

“This is literally the smallest and easiest genome to make,” Tom Ellis, a synthetic genome engineering professor at Imperial College London who was not involved in the study, told the Guardian.