Professor Jonathan Heeney did not design the vaccine the way every vaccine before it had been designed. He did not start with a known viral strain and engineer a protein to match its surface. He started with all the available genetic sequence data for an entire family of coronaviruses, logged by surveillance programs around the world, and handed the design problem to an artificial intelligence system that had never learned to think the way a virologist thinks. The system found the conserved features shared across dozens of coronavirus variants, features that evolution has preserved across strains, features that a virus cannot easily discard because they are too structurally fundamental. Then it built an antigen around those features. Not an antigen for one virus. An antigen for all of them.
On June 5, 2026, the results of the first human clinical trial of that vaccine were published in the Journal of Infection. Thirty-nine healthy volunteers received the pEVAC-PS vaccine at the University of Cambridge and its partner institution. The vaccine was safe. It generated no significant side effects. And it produced immune responses not only against SARS-CoV-2 and SARS, but against related bat coronaviruses that have never infected a human being yet.
The last word of that sentence is the point. Yet.
What Every Previous Vaccine Got Wrong
To understand what this trial means, you have to understand the structural flaw in how every COVID vaccine deployed since 2020 was designed. The flaw is not in the science. The flaw is in the assumption.
Every mRNA vaccine, every adenoviral vector vaccine, every protein subunit vaccine that went into human arms between 2020 and 2026 was designed to match the spike protein of a specific viral strain. SARS-CoV-2 as it existed when the sequence was isolated. The immune system learned to recognize that specific protein. When new variants emerged with mutations in that spike protein, Delta, Omicron, JN.1, and its successors, the vaccine’s effectiveness declined because the target had moved. The solution was boosters: updated versions of the vaccine reformulated to match the new dominant strain. Then more boosters as that strain evolved. Then more.
This is a reactive model. The virus mutates. The vaccine chases it. The virus mutates again. The vaccine chases again. The pharmaceutical infrastructure of the world has been organized, since 2020, to run faster than the virus. The race is not ending.
Heeney’s approach refuses to run the race. Instead of targeting what is on the virus’s surface, now the highly mutable spike protein, because it is under constant immune pressure, the Cambridge team used the DIOSynVax AI platform to identify features that are conserved across the sarbecovirus family. Features that appear in SARS-CoV-2, in the SARS virus that caused the 2003 outbreak, and in the bat coronaviruses circulating in animal reservoirs right now that have not yet made the jump to humans. Features that the virus cannot mutate away from without losing its ability to function at all. Those features are the target.
The AI system designed a super-antigen, a single engineered protein that presents all of those conserved features simultaneously. It is not a natural protein. No virus makes it. The AI derived it computationally from the patterns shared across the entire virus family, optimized for breadth rather than depth, for covering the whole family rather than matching any single member precisely.
The Trial That Proved It Can Enter a Human Body Safely
A Phase I clinical trial has a specific and limited purpose. It does not test whether a vaccine prevents infection. It does not establish how long protection lasts, which populations benefit most, or what dose produces the optimal response. Phase I establishes two things: the vaccine is safe enough to give to humans, and it produces a measurable immune response. Nothing more yet.
The ScienceDaily report on the trial, drawing from the University of Cambridge announcement published June 5, 2026, confirmed that 39 healthy volunteers received the pEVAC-PS vaccine through a needle-free delivery device, the PharmaJet Tropis device, which uses a microfluidic jet to deliver the vaccine intradermally without a needle. The trial was a dose escalation design, meaning participants received progressively higher doses to test the safety profile across a range.
The results: safe. Well tolerated. No significant side effects. That is the floor established by Phase I. The remarkable part of the ceiling is the breadth of immune response the vaccine produced. The trial showed that the vaccine stimulated immune responses not only against SARS-CoV-2 and SARS, but also against related bat viruses that have not yet infected humans. This is not a trivial finding. Teaching the immune system to recognize and respond to a pathogen it has never encountered, based on features shared with pathogens it might encounter in the future, is the conceptual heart of pandemic preparedness. Phase I did not prove the vaccine prevents infection by any of those viruses. It proved that the immune system responds to the antigen at all, which is the necessary first condition for protection.
The needle-free delivery is also significant, though it receives less attention. The PharmaJet device enables administration without the cold chain requirements that traditional injectable vaccines demand. Heeney noted that the DNA vaccine format chosen for this trial enables thermostability. For deployment in resource-limited settings for a pandemic preparedness tool to reach the places that are often the source of zoonotic spillover events, thermostability and needle-free delivery are not marginal conveniences. They are operational prerequisites.
What DIOSynVax Actually Built
The AI platform behind the vaccine is called DIOSynVax Digitally Immune Optimised Synthetic Vaccines. Heeney founded it as a University of Cambridge spinout in 2017 with support from Cambridge Enterprise. The platform was designed specifically for the class of problems that conventional vaccine development handles poorly: targets that are too variable, too numerous, or too unknown to address one at a time.
The design process worked from genomic surveillance data. Every time a researcher sequences a sarbecovirus from a bat in a cave in Yunnan Province, from a patient in a hospital in Wuhan, from a wastewater sample in Rotterdam, the sequence gets logged in global databases. DIOSynVax’s AI system analyzes that growing dataset and identifies the conserved regions: the structural features that appear across hundreds of sequences, the elements that different viruses in the family share regardless of how much they have diverged.
The challenge is that conserved regions are not always the most immunogenic regions. The spike protein’s receptor-binding domain is highly immunogenic; the immune system responds to it readily and powerfully, but it is also highly mutable. The conserved regions may not generate the same immediate immune response. The AI system’s job is to find the right configuration of conserved features that is both stable and immunogenic: a super-antigen that the immune system can learn from, even though it never appears in nature exactly as designed.
The pEVAC-PS vaccine delivers this super-antigen as a DNA vaccine, meaning it delivers genetic instructions rather than the protein itself. The human cell reads those instructions and produces the antigen internally, which then triggers the immune response. This format was chosen for this trial specifically because it enables thermostability and the needle-free delivery method. The full pipeline from DIOSynVax also includes candidates targeting seasonal influenza, pandemic influenza threats, and haemorrhagic fever viruses, applying the same conserved-feature approach to other families of rapidly evolving pathogens.
Why It Has Taken This Long and What Comes Next
The concept of a universal coronavirus vaccine predates COVID-19. After the 2003 SARS outbreak, researchers recognized that coronaviruses capable of human spillover existed in animal reservoirs and that the next outbreak was a matter of when, not if. The sarbecovirus family, the group that includes SARS, SARS-CoV-2, and their bat relatives, was already being studied as a family, not just as individual threats.
What was missing was the computational power to design an antigen that covered the whole family without being selected against in the way that natural viral proteins are selected against. Directed evolution and rational vaccine design could identify candidate antigens, but the combinatorial problem of finding the single configuration that maximizes both conservation and immunogenicity across dozens of sequences was computationally intractable without the kind of AI-assisted protein design that has emerged in the last decade.
The AlphaFold protein structure prediction system, developed by Google DeepMind and published in 2021, demonstrated that AI could solve structural biology problems that had resisted decades of experimental work. DIOSynVax’s approach builds on that wave of capability using AI to design rather than predict, to optimize a target rather than characterize an existing one.
Phase II trials will need to establish that the immune response demonstrated in Phase I actually translates to protection against infection. They will need to determine the optimal dose and dosing schedule. They will need to assess whether responses are durable enough to provide sustained protection or whether the vaccine requires its own periodic boosters. Phase III efficacy trials, the largest and most expensive step, will require a disease burden large enough to detect differences in infection rates between vaccinated and unvaccinated populations. For a universal pandemic vaccine, that might mean waiting for the next spillover event or running trials in populations with ongoing high-level exposure to related coronaviruses.
These are not obstacles that dismiss the Phase I result. They are the standard steps between a safety demonstration and a licensed product. Every vaccine that has ever been used in a mass vaccination campaign passed through them. The important thing about the Heeney trial is not that it finished the race. It is that it proved that the starting line is real.
The Steelman: What Phase I Does Not Prove
The honest version of the counterargument belongs in any serious assessment of this finding, because the gap between Phase I and a deployed vaccine is large and has historically swallowed more promising candidates than it has delivered.
Phase I in 39 volunteers proves safety and immune response. It does not provide protection. Immune responses measured in the laboratory, such as antibody titers and T-cell activation, are correlates of immunity, not proof of immunity. Many vaccines have generated robust laboratory-measured immune responses and failed to prevent clinical infection in Phase III trials because the correlates turned out not to be the right ones. The history of HIV vaccine development is the most prominent example: decades of candidates that generated measurable immune responses and failed to protect.
The breadth of the sarbecovirus immune response is promising, but it raises a specific scientific question that Phase I cannot answer. The conserved features the AI designed around are conserved precisely because they are structurally essential to the virus. If they are essential, they are also partially shielded from the immune access the virus has; in effect, protected those features from the immune response over evolutionary time. Whether the pEVAC-PS super-antigen can present those features in a way that the immune system can use to block infection, rather than merely recognize, is the central unanswered question.
The needle-free DNA format was chosen for operational reasons, but DNA vaccines have historically generated weaker immune responses than mRNA vaccines in human trials. Future versions of the pEVAC-PS antigen may be delivered through different platforms for different use cases. The Cambridge team stated explicitly that the super-antigen is compatible with most vaccine delivery systems. But the Phase I results reflect the DNA format specifically.
None of this diminishes the achievement. It contextualizes it accurately. A Phase I success in a universal coronavirus vaccine is genuinely significant. It is not a licensed product yet.
The Problem It Was Built to Solve
COVID-19 killed millions of people. It killed them partly because the world had not built a vaccine against the sarbecovirus family before SARS-CoV-2 emerged, despite having known for more than a decade that such a vaccine was needed. The reason it had not been built was not that the science was impossible. It was that the incentive structure of pharmaceutical development does not fund vaccines against threats that have not yet happened. Pandemic preparedness vaccines do not generate revenue until the pandemic arrives. By the time the pandemic arrives, it will be too late to develop the vaccine.
The same institutional gap that leaves developing countries receiving under 3 percent of global clean energy investment despite hosting 60 percent of the world’s best solar resources characterizes pandemic preparedness: the places most exposed to spillover risk are the places least equipped with the surveillance infrastructure, the manufacturing capacity, and the stockpiles that would contain an outbreak before it becomes a pandemic.
Heeney described what the AI platform attempts to change in a single sentence from the Cambridge press release: “We’ve converted vaccine development from being reactive to being future proof.” That sentence captures the ambition and the limitation simultaneously. Future-proof is the goal. The trial proved the approach can reach Phase I. The next steps will take years.
The cancer research community has been working on the same class of problem for a generation, universal targets that appear across cancer types, antigens that mark malignant cells regardless of which specific mutation drove transformation, and has only recently begun producing clinical results that move from laboratory promise to patient benefit. The parallel is not exact, but the timescale is instructive. Broad-spectrum approaches to biological threats require the right computational tools, the right clinical infrastructure, and years of trials that most promising early results do not survive.
The same advance in artificial intelligence that is reshaping how scientists think about protein design is now being applied to the oldest category of biological threat: the virus that spills from an animal reservoir into a human population for which no immunity exists. Heeney’s team did not wait for the next pandemic to start designing against it. They used AI to map the family first, and the vaccine second.
What the Thirty-Nine Volunteers Proved
Thirty-nine people received a vaccine designed by an artificial intelligence, delivered without a needle, targeting viruses that do not yet circulate in humans. None of them had significant side effects. All of them produced immune responses against multiple coronaviruses, including bat viruses their immune systems had never encountered in any form.
The next pandemic in the sarbecovirus family will begin the same way COVID-19 began, with a spillover from an animal reservoir into a human population that has no immunity. The question the Heeney trial is working toward answering is whether that next spillover will find a population that was already prepared, rather than a world that spends the first year of the outbreak racing to design a vaccine against a virus it has never seen.
The first human test of the AI-designed universal coronavirus vaccine found it safe and capable of teaching the immune system to recognize viruses that do not yet exist as human pathogens. That is not the end of the story. It is, for the first time, the beginning of a credible one.
