Did AI Create A Virus Found In Nature? Here's What Actually Happened

Add as a preferred source on Google

The question sounds like something out of a science-fiction trailer. But the real story, published on 6th August in the journal Science, is more precise, and arguably more unsettling, than the headline version doing the rounds online.

Scientists at Stanford University and the Arc Institute in Palo Alto did not stumble upon an AI-made virus lurking in nature. They built one. Then they built fifteen more.

Photo Credit: Canva

What The Study Did

The team, led by Stanford bioengineering graduate student Samuel King and Assistant Professor Brian Hie, trained genome language models called Evo 1 and Evo 2 on millions of natural genomes, the same way a chatbot is trained on text, except the "words" here are strands of DNA. They then prompted the model with a small, well-studied bacteriophage called ΦX174, historically significant as the first DNA genome ever fully sequenced, back in 1977.

From that single prompt, the AI generated roughly 700,000 candidate genomes. Researchers narrowed this down to around 300 of the most promising designs, synthesised them chemically, and inserted them into E. coli bacteria to see what would happen.

Sixteen of them worked.

Those sixteen produced fully functional bacteriophages: viruses that infect and kill bacteria, but cannot infect humans, animals, or plants. Some of the AI-designed phages went further than merely surviving. They replicated faster than natural viruses, and one strain overcame bacteria that had already evolved resistance to a natural phage.

Moritz Hanke, a biosecurity researcher at the Johns Hopkins Center for Health Security, described the dual-use risk plainly, noting that the same kind of model could, in theory, be asked to design a flu genome engineered for greater infectiousness rather than a harmless bacteria-killer.

So Were These Viruses "Found In Nature"?

Not quite, and this is where the study is often misunderstood. None of the sixteen viable phages existed anywhere before scientists built them. What the AI drew on was nature's patterns: millions of genomes worth of structural logic about how viral DNA is organised. It then wrote entirely new genetic blueprints that had never been seen before, some of which the researchers described as evolutionarily distant from anything known to occur naturally.

In other words, the AI didn't discover a hidden virus. It invented new ones using nature as its training material, the genomic equivalent of an AI writing an original paragraph after reading millions of books, rather than quoting one back.

Why It Matters Beyond The Lab

Bacteriophages have long been eyed as a possible alternative to antibiotics, particularly as drug-resistant infections rise worldwide. The appeal is that phages can be tailored to hunt specific bacteria, and if resistance develops, in theory a new phage could be designed to catch up. That is precisely what this study demonstrated in miniature.

But the paper itself carries an unusually candid warning about its own methods. Existing biosafety frameworks, including guidance from the World Health Organization, were not built with AI-generated organisms in mind. In the United States, a new policy introduced in July 2026 now requires computer-based (in silico) biological research to go through review if it could plausibly lead to a dangerous pathogen, with penalties for non-compliance. Whether that framework can keep pace with models like Evo 2, which is freely available on open-source platforms, remains an open question among biosecurity experts.

What We Know

The headline "AI created a virus" is technically true and easy to misread. These sixteen bacteriophages are real, lab-tested, and functional, but they are inventions built from nature's blueprint, not discoveries pulled from it. The more important story isn't whether AI can design a virus. It's that it already has, and the rules meant to govern that power are still being written.