AI Designed a Universal Coronavirus Vaccine for Viruses Not Yet in Existence

University of Cambridge researchers have developed what they describe as a “fundamentally new” type of vaccine designed exclusively using artificial intelligence (AI), marking the first time a vaccine’s active antigen was created entirely through computational modeling and tested in humans. The vaccine candidate, pEVAC-PS, was developed by DIOSynVax Ltd., a biotechnology company founded out of the University of Cambridge, England. It is a DNA vaccine using a micro fluid jet delivery system. In press materials, the vaccine candidate has also been referred to as DIOS-CoVax.^{1 2}

Unlike conventional vaccines, pEVAC-PS was designed to provide broad protection across the sarbecovirus family of coronaviruses, including all known SARS-CoV-2 variants, as well as related other bat coronaviruses that have not yet crossed into humans but carry what public health researchers identify as pandemic potential.³ The DIOSynVax team is also currently applying the same platform technology to develop separate universal flu and Ebola vaccine candidates.^{1 2}

Professor Jonathan Heeney, PhD, the Cambridge virologist and pathologist leading the research, has long expressed concern about the reactive nature of conventional vaccine development. In a recent BBC interview, Dr. Heeney said the status quo of vaccine development leaves researchers “always behind.” The goal with this technology, he told reporters, is to “finally get ahead of it.” As the first researcher to trial an AI-designed antigen in humans, Heeney said the results were “surprising” to the entire team, describing the work as “a fundamental shift in how we prepare for pandemics.”¹ He continued:

This is about making vaccines that protect us, not just from today’s viruses, but protect us from what can cause the next outbreak or disease.¹

How the First AI-Designed Vaccine Works

The first-ever AI-designed vaccine differs from conventional vaccines in a fundamental way. Typical vaccines are designed to target a specific or current strain of a virus. When a virus mutates enough, the vaccine stops working, which is why there is a new flu shot every year and why COVID-19 biologics have been updated repeatedly since 2021.¹

Rather than targeting individual viral strains, the Cambridge research team fed known genetic sequences from a broad range of coronaviruses, collected through wildlife and disease surveillance programs, into an artificial intelligence model. That AI analyzed sequence data from across the sarbecovirus family and used it to engineer what the researchers call a “super-antigen”: a synthetic antigen designed not to match any single virus, but to expose the immune system to conserved features shared across an entire coronavirus family.¹

The premise is that if the immune system learns to recognize those features, it may be able to mount a defense not only against strains that have not yet emerged in humans, but also against those with documented zoonotic spillover potential, such as H5N1 avian flu, or, in the case of the Andes hantavirus that infected passengers and crew aboard a cruise ship across multiple countries in May 2026, strains that have already made that jump.^{1 4}

Antigens are the active components of a vaccine, or the molecular markers the immune system learns to identify and attack. In conventional vaccine development, antigens are derived from known circulating virus strains and engineered by human researchers. In pEVAC-PS, the antigen was generated entirely through computer simulation, making it the first vaccine of its kind to reach human clinical trials.^{1 2}

Human Trial Participants Previously Received at Least Two COVID Shot Doses

The Phase 1 trial findings were published in June 2026 in the Journal of Infection. Notably, all of the 39 participants were between the ages of 18 and 50 and had previously received two or three doses of a COVID biologic prior to enrollment. A small population of 39 pre-injected, otherwise healthy adults cannot definitively establish efficacy, long-term safety, or how the vaccine performs across age groups, immunocompromised individuals, or people with no prior coronavirus (or SARS-CoV-2 biologic) exposure.⁵

The trial’s primary endpoints, according to its own published design, were safety and reactogenicity, not efficacy. Immunogenicity was listed as a secondary outcome, meaning the study was neither designed nor statistically powered to determine whether the vaccine generates a clinically meaningful immune response; a significant distinction when evaluating the claims being made in press coverage.⁵

Trial Authors Disclose Financial Ties to DIOSynVax and Other Trial Concerns

The follow-up window for trial participants is equally limited. According to the published trial data in the Journal of Infection, participants received an initial dose on Day 0 and a second dose on Day 28, with follow-up assessments conducted at Days 42 and 56, totaling eight weeks of observation from first injection. The study protocol allowed for visits across six months, and up to 12 months only in the event the dose was found to be highly effective.⁵

The published immune response data ends at Day 56, just short of two months. Delayed adverse events, autoimmune responses, antibody waning, and long-term durability of protection cannot be assessed within that timeframe. The participant informed consent documentation stated explicitly that pEVAC-PS had not previously been studied in humans, that potential side effects were unknown, and that “the amount of safety data available is limited.”⁵

The vaccine was developed by Cambridge’s own spin-out biotechnology company, DIOSynVax, which pioneered the AI vaccine platform technology. The researchers leading the trial have a direct financial stake in the outcome, which critics say could constitute a conflict of interest. The published trial discloses that several named authors are employees or shareholders of DIOSynVax. ⁵

AI Continues to Advance Faster Than Safety Frameworks Can Follow

AI continues to advance at speeds that have drawn intense scrutiny from tech and public health experts alike, with a growing number of critics calling for binding international agreements to ensure adequate safety and ethics measures are implemented and enforced.⁶

An extensive audit of over 2.5 million academic papers, published May 2026 in The Lancet, found that fabricated AI citations are infiltrating medical literature and may influence how doctors treat patients. Corresponding author of the audit and AI researcher Maxim Topaz, PhD, RN, based at Columbia University’s School of Nursing and Data Science Institute, described The Lancet findings as “conservative estimates,” adding, “What we identified is the lower bound of true prevalence. We’re scratching the tip of the iceberg.”⁷

AI hallucinations, the term for when AI generates false, fabricated, or biased information presented in a confident and believable manner, are documented across available large language models (LLMs). Whether analogous error rates apply to the machine learning models used in computational vaccine design has not been independently studied or disclosed by the Cambridge research team.⁷

AI Frontrunner Anthropic Calls for Global Pause in AI Development

On June 4, 2026, Anthropic PBC, the San Francisco-based AI company that develops the large language model Claude, called for a global pause in AI development, warning that current models are nearing a point where they can continue to advance without human oversight.

The Cambridge researchers have not publicly disclosed what, if any, measures were taken to ensure AI safety in the vaccine development process, including whether any mechanisms were in place to detect or correct AI-generated errors in the antigen design. The DIOSynVax platform has not been evaluated by any independent regulatory body for AI-specific safety protocols.⁶

In August 2025, U.S. Secretary of Health and Human Services Robert F. Kennedy, Jr. canceled $500 million in federal mRNA vaccine research contracts through the Biomedical Advanced Research and Development Authority (BARDA), announcing a shift toward what he described as “safer, broader vaccine platforms that remain effective even as viruses mutate.” That same year, a provision in the One Big Beautiful Bill that would have imposed a 10-year moratorium on state-level AI regulation passed the House before being stripped by the Senate in a 99-1 vote. No federal AI regulatory framework currently exists.^{8 9}

The U.S. Food and Drug Administration (FDA), the National Institutes of Health (NIH), and the Department of Health and Human Services (DHHS) have issued no public statement regarding the world’s first AI vaccine trial. No U.S. federal agency has indicated a position on AI-designed vaccine development.

If you would like to receive an e-mail notice of the most recent articles published in The Vaccine Reaction each week, click here.