The internet, being the internet, has fallen hard for a Hollywood-bound story out of Australia. Desperate man uses ChatGPT to help make a cancer vaccine for his beloved dying dog. Dog improves enough to get up off her deathbed to start chasing rabbits.
Humanity cheers and sheds a tear, both for the dog and her owner, and for the glorious future of customised AI medicine (now clearly just around the corner).
Yes, well. It is an extraordinary story to be sure, worth all of the column inches printed. But, as always, there is a little more to it.
Here are the highlights. Rosie – an eight-year-old Staffordshire Bull Terrier-Shar Pei cross, rescued from bushland outside Sydney – had something called “mast cell cancer” spreading aggressively across one of her back legs. She had perhaps six months. Her owner, Paul Conyngham, had already tried everything conventional veterinary medicine offered: multiple surgeries, chemotherapy, immunotherapy. The chemo slowed the tumours’ advance but couldn’t stop it. At that point, most dog owners would have begun the quiet process of saying goodbye.
Conyngham (who is an AI whiz but not a scientist, oncologist or biologist) opened ChatGPT and started asking questions.
What followed over the next two months – tumour DNA sequencing, protein structure modelling, a bespoke mRNA vaccine designed with publicly available AI tools, manufactured at a Sydney university and administered by researchers at the University of Queensland — has produced results that have left some of Australia’s leading oncologists effusive. Associate Professor Martin Smith, director of the Ramaciotti Centre for Genomics at the University of New South Wales, who helped sequence Rosie’s tumour DNA was one of them – “It was like holy crap, it worked. It raises the question – if we can do this for a dog, why aren’t we rolling this out to all humans with cancer?”
Conyngham’s starting point was rudimentary – he fed his problem to ChatGPT, prompting it to guide him to deeper research. The model pointed him toward genomic sequencing and immunotherapy. He used it to map out a pipeline — sequence the tumour, compare it against healthy tissue, identify the mutations driving the cancer, and then design a treatment targeting those mutations specifically. The logic is elegant: cancer cells carry genetic errors that normal cells don’t. Those errors produce aberrant proteins, called neoantigens, which are invisible to the immune system until you teach it to look. A personalised vaccine can do exactly that.
From there, Conyngham deployed Google DeepMind’s AlphaFold (the protein structure prediction tool that effectively solved a 50-year-old problem in biology) to model the three-dimensional structure of the proteins encoded by Rosie’s tumour mutations. His own machine learning algorithms then identified which mutated proteins were most likely to trigger a strong immune response. The output was a half-page formula – an mRNA sequence for a vaccine targeting Rosie’s cancer, and only Rosie’s cancer.
At this point he needed more professional help.
Prof Smith at UNSW recalled receiving what he described as one of the centre’s “oddball queries.” They reviewed Conyngham’s analysis. They were impressed enough to agree to proceed. Professor Páll Thordarson, director of the UNSW RNA Institute, designed and produced the mRNA vaccine. Within two months of receiving the genomic data, it was cold-freighted to a laboratory at the University of Queensland in Gatton, where veterinary researcher Professor Rachel Allavena had the ethical approvals to administer experimental treatments. Conyngham drove ten hours with Rosie for her first injection.
None of this emerged from a vacuum. The underlying logic – sequence a tumour, identify its unique mutations, design a targeted immune response – is now well-validated in peer-reviewed literature. A January 2026 review from Mount Sinai confirmed that personalised neoantigen vaccines are safe and capable of generating robust immune responses across melanoma, pancreatic cancer, glioblastoma, lung cancer, and bladder cancer. AI’s role in this pipeline is not decorative: tools like AlphaFold are now standard in computational oncology, and AI-driven neoantigen prediction is reducing the time from biopsy to vaccine candidate from weeks to hours.
But this is the bigger story. What Conyngham demonstrated is that this research and treatment (previously confined to well-funded research institutions and pharmaceutical giants) is now technically accessible to a motivated individual with only domain-adjacent skills and access to public AI tools. The sequencing cost him roughly $3,000. The AI tools were free or low-cost. The limiting factor was not the science; it was finding academic collaborators with the ethical approvals to proceed.
It would be easy to frame this as a story about a tech entrepreneur outwitting Big Pharma. That framing flatters the protagonist while sidestepping the real complexity. Conyngham himself is careful not to overstate. “I’m under no illusion that this is a cure,” he said. “But I do believe this treatment has bought Rosie significantly more time and quality of life.” He is already working on a second vaccine, targeting a large tumour that did not respond to the first round — which is itself instructive. Even in this remarkable case, the cancer is partially in retreat.
This is also one single story, a datapoint of one, which is not really a predictor of anything in science. A single dog, no control group, no peer-reviewed publication, no randomised trial. The scientists involved are cautiously excited, not evangelical.
The clinical bar for establishing causality is orders of magnitude higher than a jubilant dog at a park. The pharmaceutical industry’s slow, expensive, ethically constrained process exists for good reasons — most of them involving the bodies of patients who trusted compounds that hadn’t been properly tested.
And yet. The architecture Conyngham assembled – sequence, identify, model, synthesise, administer – is not meaningfully different from what companies like Moderna are doing at vast institutional scale. The AI tools he used are the same tools that cancer researchers are deploying in academic settings. His analysis was rigorous enough that UNSW scientists agreed to participate after reviewing it. The vaccine was manufactured in a legitimate university laboratory, administered by a credentialled veterinary researcher. This was not a garage operation.
What it suggests is something the pharmaceutical and regulatory establishment will need to grapple with seriously – the democratisation of the AI-assisted drug design pipeline is not a hypothetical future scenario. It has already happened, quietly, in a veterinary trial in Queensland. The question now is not whether a motivated non-specialist can execute this pipeline — Conyngham has answered that. The question is what institutions, regulators, and the broader medical community do with the knowledge that they can.
In the rush of news and commentary in the wake of this story, many scientists have expressed caution. A single case can suggest possibility; it cannot establish efficacy. Tumours sometimes shrink for reasons that are messy, partial or temporary. Treatments can coincide with improvement without fully causing it.
Of course. But there is something more interesting here – a glimpse of medicine becoming bespoke, computational and weirdly intimate.
And Rosie, for her part, is still chasing rabbits.
Steven Boykey Sidley is a professor of practice at JBS, University of Johannesburg and a partner at Bridge Capital and a columnist-at-large at Daily Maverick, Daily Friend and Currency News. His new book “It’s Mine: How the Crypto Industry is Redefining Ownership” is published by Maverick451 in SA and Legend Times Group in UK/EU, available now.
[Image: via Paul Conyngham/X]
The views of the writer are not necessarily the views of the Daily Friend or the IRR.
If you like what you have just read, support the Daily Friend