AI builds 16 brand-new viruses in world-first genome experiment
Artificial intelligence has designed 16 entirely new viruses that have never existed in nature before. Researchers in the US are calling it a genuine scientific breakthrough.
According to the team behind the project, these newly designed viruses are fully functional and can replicate successfully in lab conditions. Scientists are describing the achievement as a major turning point for the field.
This marks the first time AI has managed to design a complete viral genome from scratch. The viruses were specifically built to target bacteria, particularly drug-resistant strains of E. coli, and pose no risk to humans.
The work was led by Dr Brian Hie and his team at Stanford University. They relied on genome language models, including systems called Evo1 and Evo2, trained using data from roughly two million bacteriophages.
To keep the research safe, the team deliberately excluded any genetic data linked to viruses capable of infecting humans, animals or plants. That precaution was built into the training process from the very start.
Hie explained that this represents a new level of complexity for what generative AI can design. He noted that building a complete, functional genome capable of replicating inside cells was completely uncharted territory for his team.
Out of thousands of AI-generated designs, researchers narrowed the list down to around 300 for lab synthesis. After testing, only 16 of those proved effective at killing antibiotic-resistant E. coli.
Scientists have framed the achievement as a potential breakthrough with real medical applications ahead. Tom Ellis, a synthetic genome expert at Imperial College London, called the work impressive but pointed out its current limitations.
He noted that this particular genome remains one of the simplest and smallest to design successfully. Ellis added that concerns around AI-designed pathogens are often overstated compared to more realistic risks.
He explained that modifying existing pathogens through gain-of-function techniques remains a far easier and more likely biosecurity threat. Still, some researchers in the wider scientific community remain cautious about where this technology could eventually lead.
Experts at the Johns Hopkins Center for Health Security warned that proper oversight for this kind of AI-driven genome design doesn’t fully exist yet. Dr Filippa Lentzos from King’s College London argued that regulation needs to focus on multiple layers, including lab safety and DNA synthesis screening, rather than restricting AI models alone.
