AI Did Not Assist the Scientists Who Designed This Vaccine. AI Was the Scientist.

The University of Cambridge's AI-designed vaccine antigen is not the product of AI tools supporting human researchers. The key component was designed entirely by the AI system. That distinction is the most important one in the story.

Sign up for my Substack daily AI newsletter here.

See my AI Training course portfolio for corporate Business Leaders here.

Article link: https://open.substack.com/pub/johanosteyn/p/ai-did-not-assist-the-scientists?r=73gqa&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true

Follow me on LinkedIn: https://www.linkedin.com/in/johanosteyn/

On 5 June 2026, researchers at the University of Cambridge and its spinout DIOSynVax announced the results of the world’s first human clinical trial of a vaccine whose key component was designed entirely by artificial intelligence. The Phase 1 trial, involving 39 volunteers and published in the Journal of Infection, found the vaccine safe, well-tolerated, and capable of generating cross-reactive antibodies against multiple different coronavirus strains simultaneously — including related animal and bat coronaviruses that have not yet jumped to humans. The vaccine targets the sarbecovirus family, which includes the viruses that caused both SARS and COVID-19, as well as a range of animal coronaviruses with pandemic potential. Its ambition is to provide protection not only against all known human coronavirus variants but against bat coronaviruses that have not yet made the jump to humans and caused the outbreak that follows.

That ambition is significant. The mechanism that produced the vaccine is more significant still. Every vaccine deployed against any disease in the history of medicine was designed by human scientists working from known biological material — from the virus or bacterium or its components, selecting candidates based on human immunological reasoning and experimental iteration. The Cambridge vaccine’s active component, the antigen that trains the immune system, was not selected by a human scientist from known material. It was designed by an AI system from patterns in genetic data. That is not a faster version of the traditional vaccine development process. It is a different process — and the distinction matters for medicine, for governance, and for the boards and institutions responsible for health systems across South Africa and the world.

CONTEXT AND BACKGROUND

The problem the Cambridge team set out to solve is one of the oldest in vaccinology. Viruses mutate. When they change enough, the vaccine stops working. This is why we need a new flu shot every year, why COVID vaccines have been updated repeatedly since 2021, and why a new Ebola variant currently driving an outbreak in the Democratic Republic of Congo and Uganda is bypassing existing vaccines. The traditional response to this problem is to develop new vaccines strain by strain, updating formulations as the virus evolves and racing perpetually to stay ahead of a target that keeps moving.

AI offers a fundamentally different approach. Rather than targeting a specific strain’s proteins — which will change — an AI system can analyse the genetic sequences of hundreds or thousands of related viruses simultaneously and identify the conserved regions: the structural features that evolution has left largely unchanged across different strains, different variants, and even different species, because the virus needs them to function. Those conserved regions cannot easily mutate without the virus losing its ability to infect cells. Target them, and the vaccine should work against the whole viral family, not just the strain used as the starting point.

This is precisely what the Cambridge AI system did. It scanned the genetic sequences of the entire sarbecovirus family, including SARS-CoV-1, SARS-CoV-2 and its variants, MERS, and dozens of bat coronaviruses, identifying antigens with three properties: high conservation across strains, structural accessibility for antibody binding, and immunological potency sufficient to trigger a strong immune response. The resulting antigen — referred to in the trial documentation as the pEVAC-PS DNA vaccine — was not a modified natural protein. It was a computationally generated structure that no human scientist designed and that does not exist in nature in quite the form the AI created.

The vaccine also departs from COVID-era mRNA technology. It is a DNA vaccine — more stable than mRNA, requiring less stringent cold chain infrastructure, and delivered not by injection but by a needle-free microfluidic jet system. In low-income countries where cold chain logistics are a persistent barrier to vaccine coverage, and where needle supply and trained administration are constraints, those practical advantages matter as much as the scientific ones.

INSIGHT AND ANALYSIS

The Phase 1 trial answers three questions that every new vaccine must answer first: Is it safe? Does it produce unacceptable side effects? Does it generate the immune response it was designed to produce? The Cambridge results answer all three affirmatively. No serious adverse events were reported. The vaccine was well-tolerated across all dosing cohorts. And it generated cross-reactive antibodies — immune responses that recognise multiple different coronavirus strains, including related animal and bat coronaviruses that participants had genuinely never been exposed to. That cross-reactive immunity across strains that have not yet circulated in human populations is the defining characteristic of a future-proof universal vaccine approach, and demonstrating it in a human trial is the clinical validation that moves this from computational theory to human biology.

What the Phase 1 trial does not answer is equally important, and the researchers are explicit about it. Immune responses in this study were modest. The duration of protection is unknown. Whether further booster doses will be required is uncertain. And whether the vaccine can actually prevent coronavirus infection or reduce disease severity in the real world — the question that matters most for public health — requires Phase 2 and Phase 3 trials that are larger, longer, more expensive, and depend on viral circulation to generate meaningful endpoints. A universal coronavirus vaccine remains several years away from clinical availability. The Phase 1 results are not a finished vaccine. They are proof that the AI’s predictions about immunogenicity were correct in a human biological system — and that is the precisely right milestone for this stage of development.

The broader significance extends well beyond coronavirus. Professor Jonathan Heeney, the Cambridge scientific lead on this project, has explicitly stated that the same AI platform is currently being adapted to design universal vaccines for the Ebola virus group, seasonal influenza, and H5N1 bird flu. That roadmap transforms what might otherwise appear to be a coronavirus-specific technical advance into a platform technology — a reusable computational approach to designing broad-spectrum vaccines across multiple disease families simultaneously. The new Ebola variant driving the current outbreak in the DRC and Uganda is bypassing existing vaccines precisely because it is a different strain from the ones those vaccines were designed against. A broad-spectrum AI-designed vaccine built on conserved Ebola family regions, of the kind Professor Heeney has described, is the direct application of the Cambridge approach to that exact scenario. The Anthropic-Gates Foundation partnership announced in May 2026 — a 200 million dollar, four-year collaboration specifically targeting computationally assisted vaccine candidate screening for neglected and high-burden diseases — represents the most significant institutional bet currently being placed on AI’s potential across this entire domain.

IMPLICATIONS

The governance and institutional implications of an AI-designed vaccine passing its first human clinical trial extend well beyond the laboratory. For health systems, the question is whether the regulatory frameworks that govern vaccine approval were designed for a world in which AI can generate novel biological structures that human scientists would not have reached through conventional research. The US FDA’s regulatory pathway for a universal coronavirus vaccine making protection claims across multiple viral species is not yet defined. The question of how to evaluate the safety and efficacy of a computationally generated antigen that does not exist in nature has not been answered by any existing regulatory body. That definitional gap will need to be resolved before any AI-designed universal vaccine can reach the populations it is meant to protect.

For African health institutions and policymakers, the Cambridge breakthrough raises a question that is simultaneously urgent and underdiscussed. Bat coronavirus diversity — and therefore the zoonotic spillover risk the Cambridge vaccine is specifically designed to address — is highest in sub-Saharan Africa and Southeast Asia, the regions where pandemic preparedness infrastructure is weakest and where the consequences of the next coronavirus spillover event will be most severe. A vaccine that could provide protection against bat coronaviruses before they cause human outbreaks is directly relevant to African health security in a way that most international coverage has not acknowledged. The question for the African CDC, African health ministries, and the boards of health systems operating across the continent is whether African institutions are participating in shaping the research agenda, the clinical trial design, the regulatory frameworks, and critically, the patent architecture that will determine who this technology reaches, on what terms, and at what speed. The history of COVID vaccine distribution is not encouraging on any of those dimensions.

The DNA vaccine’s practical advantages — cold chain stability and needle-free delivery via microfluidic jet — are specifically significant for the settings where African health systems operate. That is not a coincidence of design. It is a deliberate feature of a vaccine developed with global deployment in mind. Whether that stated intention translates into equitable access at the speed and price point that African health systems require is a governance question, not a scientific one — and the answer is not guaranteed by the science alone.

CLOSING TAKEAWAY

The Cambridge AI vaccine is a genuine scientific milestone. The distinction that makes it significant — AI as the designer rather than the tool, generating a novel biological structure from computational analysis rather than selecting from known natural candidates — is the distinction that changes what is possible in vaccine development, not just how quickly it happens.

A Phase 1 trial is the beginning of a long regulatory and clinical journey. The universal vaccine it is intended to produce remains years away. But the question it has answered — whether an AI-designed antigen that does not exist in nature can generate cross-reactive immunity in humans against strains they have never encountered — is the question that had to be answered first. The answer is yes, and that changes the scope of what AI can contribute to the most consequential challenge in global public health: being ready for the next pandemic before it arrives, rather than building vaccines for it after millions have already died.

Johan Steyn is a prominent AI thought leader, speaker, and author with a deep understanding of artificial intelligence’s impact on business and society. He is passionate about ethical AI development and its role in shaping a better future. Find out more about Johan’s work at https://www.aiforbusiness.net