Cambridge Let an AI Design a Vaccine, Then Tested It in 39 People

Welcome back to SavvyMonk, your one-stop for AI and tech news that actually matters.

Today a team in the UK did something that has never been done before, letting an AI design the working part of a vaccine and then putting that design into real people. The result is more interesting for what it proves about the method than for the numbers it produced, which is exactly why it is worth your time.

Let's get into it.

You know the fix: longer prompts, more context, clearer constraints. But typing all that takes five minutes per prompt, so you shortcut it. Every time.

Wispr Flow lets you speak your prompts instead of typing them. Talk through your thinking naturally — include context, constraints, examples — and get clean text ready to paste. No filler words. No cleanup.

Works inside ChatGPT, Claude, Cursor, Windsurf, and every other AI tool. System-level, so there's nothing to install per app. Tap and talk.

Millions of users worldwide. Teams at OpenAI, Vercel, and Clay use Flow daily. Free on Mac, Windows, and iPhone.

Try Wispr Flow free

Researchers at the University of Cambridge and its spin-out company DIOSynVax have run the first human trial of a vaccine whose active ingredient was designed entirely by artificial intelligence, a milestone the team reported through Cambridge and described as a world first.

The shot, called pEVAC-PS, is built around what the scientists call a super antigen, a single synthetic target that an AI assembled after scanning the genetic blueprints of thousands of coronaviruses. For decades vaccine antigens have been chosen and shaped by people working at a bench, so handing that core design task to a machine and then testing the output in human arms marks a real shift in how this work gets done.

The early-stage study enrolled 39 healthy volunteers between the ages of 18 and 50, all of whom had already received two or three doses of an existing COVID vaccine. It ran at NIHR clinical research facilities in Southampton and Cambridge, according to the hospital that hosted part of it, and it was designed first to check whether the vaccine was safe rather than to prove it works at scale.

How a Machine Designed the Antigen

The approach flips the usual logic of vaccine design. Every licensed COVID vaccine so far has been engineered around specific strains that were already circulating, which is why the shots need regular updating as the virus drifts.

The Cambridge team instead fed an AI system the genetic sequence data that surveillance programmes around the world have logged for sarbecoviruses, the broad family that includes SARS-CoV-2 and the original SARS virus along with related coronaviruses still sitting in bats. The model then built one antigen aimed at the features all of those viruses share.

Those shared features matter because they tend to be the parts a virus cannot easily change without breaking itself. Scientific lead Jonathan Heeney, Professor of Comparative Pathology at Cambridge and the founder of DIOSynVax, has said the design targets the elements that are essential to the virus staying alive, which is what limits the room a future strain has to mutate its way around the shot.

He framed the wider goal as moving vaccine development away from something reactive and toward something closer to future proof, and he compared the usual habit of chasing each new variant and rushing out an updated shot to a dog chasing its own tail.

What the Trial Actually Showed

This is where the honest reading matters. The vaccine was safe and well tolerated across all four escalating dose levels, given on the first day and again 28 days later through a needle-free jet rather than a syringe, and the researchers recorded no significant safety concerns. That is the headline result, and for a brand new design method it is the result that counts most.

The immune response, though, was modest. Because nearly every volunteer already carried strong immunity from earlier COVID shots, and because fresh waves of SARS-CoV-2 kept washing through during the recruitment window, it was hard for the trial to show a dramatic jump in protection on top of what people already had.

The team did measure real responses, including reactions to SARS-CoV-2, to the original SARS virus, and to related bat coronaviruses that have never infected humans, which is the breadth they were hoping for. But anyone reading the early coverage as proof that an AI has cracked the universal vaccine is getting ahead of the data, because what the trial proves is that the method is safe enough to keep going, not that it is ready for anyone's arm outside a study.

Why a Universal Shot Is the Prize

The reason this line of work draws so much money and attention is the prize at the end of it. A shot that protects against an entire family of viruses, including ones that have not yet jumped from animals into people, would change how the world prepares for the next outbreak.

Saul Faust, the chief investigator on the trial, has pointed out that a universal vaccine could in principle guard against several variants at once and even against related viruses that have not yet emerged in humans, which is the difference between bracing for a known threat and being ready for an unknown one.

DIOSynVax has been building toward this for a while. The company spun out of Cambridge in 2017 with the name Digitally Immune Optimised Synthetic Vaccines, and its pipeline runs well beyond coronaviruses into seasonal and pandemic influenza and haemorrhagic fever viruses.

The pandemic preparedness group CEPI has backed Heeney's broader coronavirus work with a five year grant worth 42 million dollars, a sign that serious institutions see the platform as more than a one-off experiment.

The Double Edge of AI in Biology

There is a sharper context worth holding alongside the good news. The same week this trial surfaced, the chiefs of OpenAI, Anthropic, Google DeepMind, and Microsoft AI signed a public letter warning Congress that AI is eroding the knowledge barriers that once kept dangerous biology out of reach, and pushing for mandatory screening of synthetic DNA orders.

The tools that let a model design a protective antigen are close cousins of the tools that could design something harmful, and the field is starting to grapple with both ends of that range at the same time. A vaccine you can build in software and ship as DNA is a triumph and a warning rolled into one, and the people closest to the work are the ones saying so most clearly.

What Comes Next

The next step is a larger Phase 2 trial in a more diverse group of people, designed to find out whether the vaccine can generate the strong, broad protection it is built for rather than only proving it is safe. That study has to clear a higher bar, and it will take time.

None of this means a universal coronavirus shot arrives soon, and the gap between a clean Phase 1 and an approved product is where most promising vaccines stall. What is different here is the engine behind the design, because if an AI-built antigen can carry a vaccine through the full process, the same approach can be aimed at the next family of viruses long before it becomes a crisis.

The Bottom Line

The story worth keeping is the method, not the numbers. An AI designed the working core of a vaccine, that vaccine went into people, and it came out the other side safe, which is the part that has never happened before.

The modest immune response is a reminder that proof of concept is not proof of protection, and the real verdict waits on a bigger Phase 2. Watch that trial, and watch the parallel push to screen synthetic DNA, because the same toolkit that designed this shot is the one the industry is now asking governments to keep an eye on.

Category: Research

"Act as a skeptical science analyst. I will paste a press release or news story about a breakthrough in [field]. Separate what was actually demonstrated from what is being claimed or implied. List the hard results, the limitations the study itself admits, the sample size and stage, and the three questions I should ask before I believe the headline. Finish with one sentence on what evidence would change your assessment."

For most of modern medicine, the slow part was a person sitting with a problem until the right design finally appeared. Watching a machine do that first pass, and then watching real volunteers test the output, is a glimpse of how much of science is about to speed up.

The catch is that faster design cuts both ways, and the same week that gave us an AI-made vaccine also gave us four AI chiefs asking for tighter rules on the very tools behind it. Hit reply, I read every response.

See you in the next one.

— Vivek

P.S. If you know a researcher, founder, or anyone tracking where AI and biology collide, forward this their way. They can subscribe at https://savvymonk.beehiiv.com/