Two AI-designed vaccines enter human trials; Pfizer and Mayo Clinic bet big on AI
Cambridge's pan-coronavirus vaccine and Oxford's CCHF vaccine clear first human trials as Pfizer licenses Chai-3 and Mayo Clinic partners with Microsoft on a clinical AI model.
By BINA Editorial
Historic double milestone: AI-designed vaccines reach human testing
Two vaccines designed with artificial intelligence reached human clinical trials this week, marking what researchers are calling a turning point in how medicine is developed.
Researchers at the University of Cambridge and spin-out company DIOSynVax announced that their pan-Sarbeco coronavirus vaccine completed a Phase I study in 39 healthy volunteers. Unlike conventional shots, the vaccine was designed by an AI system to target structural features shared across the entire Sarbecovirus family — the group that includes SARS, SARS-CoV-2, and related bat coronaviruses. The AI identified conserved regions that human immune systems rarely encounter on their own, aiming for broad protection against future outbreaks. The trial found the vaccine safe and well-tolerated.
On the same day, the University of Oxford and biotechnology company Basecamp Research announced a separate first-in-human trial for a vaccine against Crimean-Congo haemorrhagic fever (CCHF), a tick-borne virus with fatality rates reaching 40 percent and no approved vaccine. The CCHF vaccine was also engineered through an AI-driven biological discovery pipeline — making the week's double announcement the first documented instance of two AI-discovered vaccines entering human trials simultaneously. Both programmes will report preliminary immunogenicity data later this year.
Pfizer licenses Chai-3 for antibody discovery
Pfizer signed a licensing agreement with AI drug-discovery startup Chai Discovery, granting access to Chai-3, the company's newest model. Chai says Chai-3 doubles the success rate of antibody design compared with its predecessor and can target difficult-to-drug molecules by predicting their three-dimensional structure with higher accuracy. Pfizer will also receive a custom model tuned on its own proprietary data and workflows, aimed at shortening the time from target identification to clinical candidate.
The deal follows a wave of similar pharma-AI tie-ups — Sanofi and Owkin, Alnylam and Inceptive — underscoring how quickly generative biology has moved from academic proof-of-concept to the drug pipeline.
Mayo Clinic and Microsoft to build a healthcare-specific AI frontier model
Mayo Clinic and Microsoft announced a collaboration to build what they are calling a "frontier model" for clinical medicine. The project will combine Mayo's de-identified patient records — one of the world's largest clinical datasets — with Microsoft's cloud and engineering resources. Mayo will own the resulting model; Microsoft plans to make it available to other health organisations through its Azure AI Foundry platform.
Developers say the model will be trained on multimodal data including physician notes, imaging, lab results, and genomics, with the goal of supporting earlier and more accurate diagnoses and treatment planning. The partnership reflects a broader shift in which health systems are moving beyond licensing generic large language models toward building domain-specific foundation models on proprietary clinical data.