Within two months of completing genetic sequencing, Rosie received her first injection in December 2025, and by mid-March, a tennis ball-sized tumour on her leg had shrunk by roughly 75%. For a dog whose mobility had deteriorated in December, Rosie was jumping over a fence to chase rabbits by late January. The story has circulated widely as evidence that artificial intelligence can revolutionise cancer treatment.
Yet the research reveals something more measured and more important than the headlines suggest. The treatment worked for one tumour in one dog, and understanding the distinction between partial response and cure matters enormously to anyone considering these vaccines as cancer therapy.
What the data actually shows
Rosie was diagnosed with advanced mast cell cancer in 2024, and chemotherapy slowed the spread but couldn't shrink her tumours. Conyngham spent $3,000 to have Rosie's healthy DNA and tumor DNA sequenced at the University of New South Wales, then used AI tools including AlphaFold to pinpoint the mutations driving her cancer and identify potential drug targets.
Conyngham brought his data to UNSW's Ramaciotti Centre for Genomics, where researchers initially hesitated but agreed to collaborate after reviewing his analysis. Pall Thordarson, director of the UNSW RNA Institute, designed and produced the bespoke mRNA vaccine, and it was less than two months from when Conyngham designed the sequence until the construct was handed over.
Clinical significance here is critical. One of Rosie's tumours didn't respond, and the team is already sequencing it to design a second vaccine. UNSW published a blog post specifying that Rosie still has cancer and it is still incurable, though she was doing better and her tumours have shrunk so much you can see her legs again. This is not a cured dog; it is a dog with better symptom control from one partially successful treatment.
What AI actually did versus what headlines claim
The framing of this story in social media and some news outlets suggests AI designed the vaccine. AI tools assisted with research and data exploration, but did not design the cancer therapy, despite headlines saying so.
ChatGPT served as a research assistant and planning tool, AlphaFold predicted protein structures, but the critical decisions—which neoantigens to target, how to design the mRNA sequence, and how to confirm the design—required human judgment from Conyngham and deep domain expertise from Thordarson's team at UNSW. The AlphaFold confidence score for Rosie's c-KIT protein was 54.55, which UNSW structural biologist Dr. Kate Michie publicly described as low, noting that AlphaFold can get things wrong and significant lab work is needed to validate any output.
The AI accelerated a pipeline that would otherwise require months of manual literature review and computational modelling. That acceleration is genuinely significant. Compressing research that might take months into a timeline of weeks changes what becomes possible for someone with technical expertise and institutional access. But this is different from saying AI designed a cancer cure.
The context investors and patients should understand
This is one dog, one tumour, with no controlled trial. Mast cell tumours can sometimes shrink spontaneously. Human trials would require years of regulatory work and hundreds of millions of dollars in testing.
Rosie's treatment was possible because veterinary experimental treatments face lighter regulatory scrutiny than human medicine. There is no veterinary equivalent of FDA Phase I-III clinical trials for a one-off compassionate use case. Scaling this to a standardised treatment for dogs or humans would require years of regulatory work that this single case does not shortcut.
The underlying science is sound. The same mRNA platform technology is already being tested in dozens of human clinical trials by some of the biggest names in pharma, and the most advanced programmes are nearing potential regulatory approval. Moderna and Merck's jointly developed personalized melanoma vaccine showed a 49% reduction in the risk of cancer recurrence or death over five years of follow-up when combined with Merck's Keytruda.
What Rosie's case demonstrates is that the pipeline—sequence the tumour, model the mutations, design a targeted mRNA vaccine—can be compressed from years into months using AI tools that are freely available today. That is the headline worth reporting.
What patients should take from this
Conyngham himself is clear-eyed about the limits: I'm under no illusion that this is a cure, but I do believe this treatment has bought Rosie significantly more time and quality of life. Improved quality of life and extended survival are not trivial goods in cancer treatment, and they matter clinically. Neither is a cure, and the distinction shapes reasonable expectations.
The research shows that personalised medicine using mRNA technology can be effective when designed properly and administered under careful conditions. What it does not show is that AI systems have solved cancer or that regulatory pathways are unnecessary obstacles. Both propositions have circulated in social media responses to this story, and both distort what the evidence actually tells us.
Rosie's improved mobility and quality of life are genuine achievements that matter to her owner and to researchers developing human treatments. The speed with which her treatment was designed and manufactured signals real progress in making personalised cancer vaccines faster and cheaper to produce. But distinguishing between those concrete findings and the broader claim that AI has fundamentally changed cancer treatment is where evidence-based health reporting must hold the line.