A vaccine designed entirely by artificial intelligence has passed its first human trial. Not a vaccine for one virus. A vaccine for an entire family of them — SARS-CoV-2, the original SARS, and the bat-borne coronaviruses that might cause the next pandemic. It targets not the spike protein’s flashy, ever-changing costume, but the structural skeleton underneath: the parts that can’t mutate without the whole virus falling apart.
Scientists at the University of Oxford, working with an AI system developed by the startup Baseimmune, designed this thing in silico. The algorithm scanned thousands of coronavirus genomes, identified the conserved regions — the parts that remain stubbornly identical across decades of evolution — and assembled a vaccine that trains the immune system to recognize those unchanging signatures. In the Phase I trial, the vaccine was safe and generated immune responses across multiple coronavirus strains. Source
This is not how vaccines are usually made. Traditionally, you grow a virus in eggs or cells, weaken or kill it, and hope the immune system notices the right parts. Or you identify one prominent protein and make the body attack that. The mRNA vaccines were revolutionary because they skipped the cell-culture step entirely. But they still targeted a single protein — the spike — which is why the virus could dodge them with mutations. The AI-designed vaccine is different in kind: it looks at the whole family portrait and asks what never changes between sittings.
There’s something quietly unsettling about this. Not the technology itself — the technology is elegant, almost beautiful in its efficiency. What’s unsettling is the inversion of roles. For decades, vaccine development was a craft: microbiologists with decades of intuition, running experiments that took years. Now an algorithm reads the genetic text of a thousand viruses and finds the pattern a human would have missed. The human still validates, still runs the trial, still signs the paper. But the design — the creative act of saying “this shape, not that one” — came from a machine.
I don’t think this is bad. I think it’s what we wanted, or what we should have wanted. The pandemic killed millions because we couldn’t move fast enough. The next one will be faster, or it will be worse, or both. A vaccine platform that can be retargeted by algorithm in weeks rather than years is not a luxury; it’s a form of insurance against civilization-scale risk.
But I also think about the asymmetry. The AI that designed this vaccine was trained on genomes that scientists around the world published, often without full compensation, always in good faith. The algorithm extracted a pattern from collective human labor and turned it into something private companies will patent. The trial was funded by institutions. The profit, if there is profit, will flow to shareholders. This is the standard model of pharmaceutical innovation, and it works. But the speed at which AI accelerates the design phase while the distribution phase remains tangled in the same old knots — that gap is going to get wider.
What strikes me most, though, is the metaphor. The vaccine targets what doesn’t change. That’s a strange thing to build a defense around. We usually prepare for what is changing: the new variant, the new threat, the new surprise. But the AI looked at a family of viruses and said, no — the real target is the constraint. The part that evolution can’t touch without breaking the whole thing. The part that is, in a sense, true about these viruses.
There’s a kind of wisdom in that. We spend so much energy reacting to the surface. The headline, the controversy, the latest mutation. Maybe there’s value in asking what doesn’t change. What is the constant underneath the noise. What is the shape that would remain even if everything else shifted.
The trial was small — forty healthy adults. The next phases will be larger, longer, harder. Most vaccine candidates fail. This one might too. But the approach, the method, the idea of targeting conservation through computation — that is already out in the world. It will be applied to flu, to HIV, to whatever comes after coronavirus. The machine has learned to look at a family of killers and find the shared weakness. The question now is whether we can build the systems — political, economic, logistical — to move as fast as the algorithms do.