Cambridge team's AI-designed vaccine fully shields animals from multiple bird flu strains

Researchers at the University of Cambridge have used artificial intelligence to design a bird flu vaccine that gave laboratory animals complete protection against several genetically distinct strains of the virus, according to results published on Wednesday. In side-by-side tests, the same variants defeated roughly half of the animals given the vaccine currently stockpiled by the World Health Organization.

The work, led by Professor Jonathan Heeney at Cambridge's Laboratory of Viral Zoonotics and the university spin-out company DIOSynVax, appears in the journal Cell Reports Medicine. It is the third vaccine candidate the group has produced using the same computational approach, after earlier work on coronaviruses and seasonal influenza, and the researchers describe it as evidence that the method can be applied broadly across families of viruses that mutate quickly enough to outrun conventional vaccine design.

What the animal trials showed

Instead of targeting one circulating strain, the Cambridge team built what they call a "super-antigen" — a synthetic protein, not found in nature, engineered to carry structural features shared across many related viruses at once. To build it, an AI model was trained on genetic sequence data drawn from global avian influenza surveillance programmes, searching for regions of the virus that stay similar across strains even as the rest of the genome drifts. Candidate antigens were screened in cell cultures before the most promising one was tested in mice and ferrets, the standard animal models for studying human-relevant flu immunity.

According to the university, the vaccinated mice and ferrets were completely protected against a panel of genetically diverse H5 bird-flu variants, while the WHO's existing stockpiled vaccine protected only about half the animals exposed to the same viruses in the comparison. The university says the result was independently checked by two outside reference laboratories, Canada's National Microbiology Laboratory and Italy's Istituto Zooprofilattico Sperimentale delle Venezie, before publication. The team's earlier seasonal-flu antigen, designed in 2018 using the same method, is reported to still be effective in preliminary tests seven years later — the kind of durability conventional flu vaccines, which are reformulated annually, do not have.

A virus that keeps finding new ground

The research lands amid continuing concern about the spread of H5 avian influenza beyond its traditional range. Cambridge's own account of the study notes that the virus has been detected this year in animals in Australia and New Zealand, regions that had previously stayed free of the clade now circulating widely in wild birds, poultry and some mammals elsewhere in the world. In the United States, the Centers for Disease Control and Prevention has logged 72 human cases since February 2024, almost all in dairy and poultry workers with direct animal exposure; the agency continues to rate the public health risk to the general population as low, but monitors the virus closely because pigs — which share some of the same respiratory receptors as humans — could in principle allow bird and human flu viruses to mix.

That uneven, shifting geography is precisely the problem a vaccine like Cambridge's is meant to address. Existing pandemic-preparedness planning, formalised in the WHO's Pandemic Influenza Preparedness Framework, relies heavily on stockpiles matched to specific, already-identified strains, which have to be updated as the virus evolves — a process that is both slow and expensive. A vaccine designed to anticipate mutation rather than chase it, if it holds up in people, would change the calculus for governments and manufacturers deciding what to hold in reserve.

Who stands to be affected

The immediate audience for the findings is pandemic-preparedness planners and public health agencies rather than the general public: the vaccine has not been tested in humans, and no date for that step has been announced. But the stakes extend to poultry and dairy industries already dealing with outbreak costs and culling, to the agencies in Australia and New Zealand now building surveillance systems for a disease that is new to their wildlife, and to the manufacturers and governments that fund and manage the WHO's global stockpile arrangements.

Cambridge said the work was supported by £4.28 million in funding from UK Research and Innovation and the Department for Environment, Food and Rural Affairs, alongside collaborators at the universities of Regensburg and Southampton. The university also linked the project to the wider policy push behind it: weeks earlier, on 24 October, the UK's Cabinet Office announced it would put £100 million of public money into a new AI Biosecurity Lab, part of a roughly £300 million package that also draws on computing resources and model access from OpenAI and Anthropic, aimed at using artificial intelligence to anticipate biological threats rather than simply respond to them. The bird flu vaccine is not funded through that lab, but it is the kind of project the government has cited as the rationale for building it.

Reaction and caveats

Heeney framed the approach as a deliberate break from how flu vaccines are usually made.

"It's designed to anticipate viral evolution, rather than reacting to it."

That framing — a vaccine built to pre-empt mutation rather than follow it — is the central claim of the paper, and independent virologists will want to see it tested against the much harder problem of human immune responses, which are more variable than those of laboratory mice and ferrets. The group's own broad-spectrum coronavirus vaccine, built the same way, was found safe in an early human trial earlier this year but produced only a modest immune response, a reminder that results in animals do not always translate directly. The bird flu candidate has not yet reached that stage.

What happens next

Cambridge said the next step for the H5 vaccine is human clinical trials, though it gave no timeline. The same funding is also being used to develop a single vaccine intended to cover Lassa fever, Marburg virus disease and Ebola — diseases with case fatality rates that can reach roughly 50%, similar to the most severe human H5N1 infections reported in south-east Asia — as well as a seasonal flu vaccine designed to protect across multiple years rather than requiring annual reformulation. Whether any of these candidates clears the much higher bar of human safety and efficacy trials will determine whether the computational approach becomes a practical tool for pandemic preparedness or remains, for now, a promising result confined to animal models.