Stanford Student Used Generative AI to Design Virus Genomes, and the Field Is Still Arguing Over What It Means

In 2025, Samuel King, then a PhD student at Stanford University, ran a generative AI model on a problem that had mostly belonged to biology labs: writing genetic sequences for microscopic viruses. The model returned blueprints, not organisms. The distinction matters, but so does the direction of travel — the work is widely treated as a step toward AI-designed life, even if nobody is calling it that yet.

The technical gulf between a sequence and a functioning virus is where most of the debate sits. A generative model can propose a plausible genome; it cannot by itself assemble the proteins, package the genetic material, or demonstrate that the resulting particle infects anything. King's output was a candidate design, and the question of whether such designs can be synthesized and tested is now a practical one rather than a hypothetical.

That gap is closing fast. DNA synthesis has become cheap and widely available, and the screening regimes that are supposed to catch dangerous orders are uneven across suppliers and jurisdictions. A model that proposes viral sequences is only half the story; the other half is the commercial infrastructure that turns strings of letters into physical material, often with limited verification.

The research also sits inside a broader shift in how biology gets done. Generative models trained on protein and genomic data are already used to propose new enzymes, antibodies and scaffolds, and labs have shown they can produce working designs in some cases. Viruses are a harder test because they are not just molecules — they are self-replicating systems, and the design problem includes packaging, host range and stability.

For biosecurity researchers, the concern is less about any single experiment than about the normalization of the capability. Once a method is published and a model is available, the cost of reproducing it drops for everyone, including people with bad intentions. That has pushed calls for better DNA synthesis screening, clearer norms for publishing viral design work, and more deliberate review before results go out.

What happens next likely hinges on two things: whether King's designs, or similar ones, are actually built and tested in a controlled lab, and whether the field settles on rules before that happens. The answer will shape how much of this work stays in universities and how much moves somewhere with fewer guardrails.

MIT Technology Review is hosting a conversation with King to dig into the experiment and its implications. The interesting question is not whether AI has designed life — it hasn't, yet — but how quickly the remaining steps get taken.