AI Creates First Functional Viruses
Stanford scientists used genomic language models to design bacteriophages that destroy superbugs, but biosecurity experts warn that legal frameworks are not keeping pace with the technology.
Stanford scientists used genomic language models to design bacteriophages that destroy superbugs, but biosecurity experts warn that legal frameworks are not keeping pace with the technology.
In a historic leap for biotechnology, a research team led by Dr. Brian Hie at Stanford University has, for the first time, used artificial intelligence to create entirely new, functional viruses. While the very idea of AI-generated viruses may cause unease, these are bacteriophages-viruses that target only bacteria and are completely harmless to humans, animals, and plants.
The research findings, published on August 9, 2026, in the prestigious journal Science, open the door to revolutionary applications in the fight against antibiotic-resistant bacteria, but they also serve as a serious warning about the lack of safety mechanisms to control this technology.
At the heart of the discovery are models called Evo1 and Evo2, described as the genetic equivalent of the advanced AI systems that power today's chatbots. The key difference is that these models do not learn words and sentences; they read and generate complex genetic codes. They were trained on a massive dataset of two million natural bacteriophages, with scientists deliberately excluding genes from viruses dangerous to humans and animals to minimize risk.
After the AI generated thousands of potential genomes, the researchers selected 300 of them for synthesis in the lab. These artificial genetic codes were inserted into bacteria, prompting the cells to produce real, live viruses. Although the success rate was relatively low-only 16 bacteriophages survived out of all attempts-the final result exceeded expectations.
Laboratory tests demonstrated the remarkable potential of the newly created viruses. A combination of these AI-designed bacteriophages quickly and effectively destroyed strains of E. coli bacteria that had previously developed resistance to natural phages. This discovery could fundamentally change phage therapy, a therapeutic method gaining increasing attention in treating serious infections caused by superbugs.
The ability to rapidly redesign a viral genome in response to specific bacterial mutations gives medicine an incredibly flexible weapon. Instead of the lengthy process of searching for natural phages, AI could create customized viruses for each patient in real time.
The scientific triumph, however, is accompanied by serious concern. Biosecurity experts from the Johns Hopkins Center in Baltimore emphasize an alarming gap: while the technology for creating viral genomes using generative AI already exists and works, the legal and safety frameworks for its control are still in their infancy.
Although bacteriophages, due to their extremely small genomes, are just the first and simplest step, there is justified fear of misuse. What if similar models were trained on pathogens dangerous to humans? That is why experts are urging that rigorous controls not focus solely on the AI software level, but primarily on the companies that physically synthesize and produce ordered DNA chains.
The Stanford discovery undeniably proves that artificial intelligence can now design independent life at the molecular level. This fact simultaneously opens the door to new medicines and creates a race against time to establish global biosecurity rules before the technology goes too far.
