Story No. 001 · Medicine · Australia
The Man Who Used ChatGPT to Build a Cancer Vaccine for His Dog
Paul Conyngham had no medical background. His dog Rosie had terminal cancer and months to live. So he opened a chat window — and didn’t close it for two years.
In June 2023, Paul Conyngham noticed something wrong with Rosie. Swollen lumps on her head and leg. He took her to the vet. Twice, they told him not to worry — probably just a rash. The third visit produced a biopsy. The result arrived in May 2024: aggressive mast cell cancer. Terminal. A few months to live.
Conyngham, a Sydney-based tech entrepreneur and AI consultant, tried everything conventional medicine offered. Chemotherapy. Surgery. Standard immunotherapy. The tumours kept growing. Costs were mounting.
Most people, at this point, would have started preparing for the worst.
Conyngham opened ChatGPT instead.
A Normal Dog Walk
That Changed Everything
“I would have conversations and just keep them going non-stop,” he later said. Not casual questions. Deep, hours-long research sessions across ChatGPT, Gemini, and Grok — building a working understanding of cancer biology, emerging treatments, mRNA vaccine technology, and genomics from scratch. He wasn’t skimming. He was studying.
The turning point came in August 2025, more than a year into his research. Two earlier approaches had already failed. He had screened over a million existing compounds looking for a molecular match. He had tried to license a patented drug whose holder refused compassionate use.
Then, in a late-night conversation with ChatGPT, a new direction emerged: could he design a personalised mRNA vaccine himself?
The idea is scientifically grounded. mRNA vaccines teach the immune system to recognise and attack a specific target — in cancer treatment, the unique protein signature of a patient’s tumour. A neoantigen vaccine is built around that exact signature. The same technology powered the COVID-19 vaccines. Researchers have been exploring whether the same speed could be applied to cancer for years.
“The chatbots empowered me as an individual to act with the power of a research institute — planning, education, troubleshooting, compliance, and yes, real scientific design work.”Paul Conyngham — post on X, March 2026
From Chat Window
to Research Lab
What Conyngham did next was methodical. Following a recommendation from ChatGPT — including a specific suggestion to contact researchers at the University of New South Wales — he paid $3,000 to have both Rosie’s healthy genome and her tumour genome sequenced at the Ramaciotti Centre for Genomics at UNSW Sydney.
He used AI systems to identify a target protein and an FDA-approved substance that could support the treatment. The final vaccine design was created using the Grok AI model. A referral chain — built through ChatGPT’s suggestions — led him to Professor Pall Thordarson at the UNSW RNA Institute, who manufactured the mRNA. The vaccine was administered at the University of Queensland’s School of Veterinary Science. AI also helped Conyngham navigate the ethics approval paperwork.
Three weeks after the first treatment, Rosie’s tumours swelled dramatically. Conyngham was alarmed. Researchers recognised it immediately: pseudoprogression. The immune system was sending T-cells directly to the tumour. The swelling was evidence it was working exactly as designed.
The reduction in size of Rosie’s largest tumour three months after the experimental mRNA vaccine and immunotherapy protocol began
Paul Conyngham via AFP · March 2026
Two vet visits dismiss the lumps as a rash. The misdiagnosis delays treatment by nearly a year as Rosie continues to deteriorate. By the third visit, the lumps are no longer dismissible.
Chemotherapy, surgery, and standard immunotherapy all attempted and failed. Costs mount. Conventional medicine has run out of options. Conyngham opens ChatGPT.
Over a million compounds screened. A patented drug’s compassionate use denied. Then in conversation with ChatGPT, the concept of a personalised neoantigen mRNA vaccine emerges as a viable direction.
Both of Rosie’s genomes are sequenced at UNSW for $3,000. AI identifies the target protein. The mRNA sequence is designed with Grok. Professor Thordarson’s lab manufactures it. Ethics approval navigated with AI assistance.
Three weeks after treatment, tumour areas swell. Conyngham panics. Researchers recognise pseudoprogression: T-cells flooding the tumour exactly as the vaccine was designed to trigger. The swelling is evidence it’s working.
Rosie’s largest tumour has shrunk ~75%. Two areas appear normal. Conyngham publishes his process as open-source and opens a Google form for other dog owners facing similar diagnoses.
The Story That
Split the Internet
Sam Altman shared the story on X. OpenAI President Greg Brockman shared it. DeepMind CEO Demis Hassabis shared it. Altman’s post reached 1.3 million views. Tech Twitter was moved. But not everyone was.
Egan Peltan, a Stanford-trained PhD in chemical biology and co-founder of a biotech startup, pushed back sharply. His core argument: there is zero controlled evidence the AI-designed vaccine did anything. Rosie was receiving conventional immunotherapy at the same time. The improvement is most likely a response to that drug, not the mRNA vaccine. True cost of treatment: $20,000 to $50,000. His verdict: “storytelling for AGI true believers.”
Both reactions are reasonable. That tension — between the genuine emotional power of this story and the legitimate scientific caution it deserves — is exactly what makes it worth sitting with.
“Paul didn’t have a biology degree. He had 17 years of pattern recognition, a dying dog he loved, and the willingness to treat an impossible problem as a data problem.”— Comment on Conyngham’s LinkedIn post, widely shared
Conyngham didn’t replace scientists. He connected them. He identified the right researchers, asked the right questions, designed the protocol, and navigated the ethics paperwork — work that would have taken a research institution years to initiate. AI gave him the intellectual scaffold of a team while working alone from a laptop.
UNSW’s Professor Martin Smith: “The combination of three disruptive technologies — genome sequencing, artificial intelligence, and RNA therapeutics — offers new possibilities and challenges.” The possibility here is that personalised cancer treatment, currently confined to expensive clinical trials, could one day become accessible to anyone with enough determination and the right tools.
Rosie was receiving conventional immunotherapy at the same time. The improvement is most likely a response to that drug — not the mRNA vaccine. Egan Peltan, Stanford-trained PhD in chemical biology: “There is zero evidence that the AI-assisted work did anything for Rosie’s cancer.”
Nick Semenkovich at the Medical College of Wisconsin: “We don’t know enough about the vaccine to understand how much AI helped — or if the vaccine worked the way it was designed.” Personalised mRNA cancer vaccines have been in development for years. The field needs Phase 3 results. True cost: $20,000–$50,000.
“It was really driven by his determination to help his dog. This was not a clinical trial by any means. It’s not that AI cured cancer.”
“AI transforms a needle-in-a-haystack search into a data-driven selection process, drastically shortening the timeframe between diagnosis and vaccine construction.”
“UNSW and Conyngham haven’t published scientific details outside of interviews, so we don’t know enough to understand how much AI helped — or if the vaccine worked the way it was designed.”
“The coolest meeting I had this week was with Paul, who used ChatGPT and other LLMs to create an mRNA vaccine protocol to save his dog Rosie. It is an amazing story.”
What is a personalised mRNA cancer vaccine?
mRNA vaccines work by delivering genetic instructions to your cells, teaching them to produce a specific protein that the immune system then learns to recognise and attack. In cancer treatment, the target protein is derived from the unique mutations in a patient’s own tumour — called neoantigens.
Because every tumour is genetically distinct, a neoantigen vaccine is personalised to that specific cancer profile, not a general treatment. The approach was proven possible during COVID-19. Researchers have since been exploring whether the same speed and personalisation could be applied to cancer — in humans and, as Rosie’s case demonstrates, in animals.
What did each AI tool actually do?
ChatGPT was the primary research partner — deep literature review, treatment strategy, troubleshooting, navigating ethics approvals, and connecting Conyngham to the UNSW research team.
Gemini was used for cross-referencing. Grok was used specifically for the final vaccine design — converting genomic data into an mRNA sequence prescription. AlphaFold was used to analyse relevant protein structures. All tools worked alongside human experts at every stage.
Has Conyngham shared his process with others?
Yes — in full. He published a detailed account on X covering each stage from initial research through genomic sequencing to vaccine design and administration. He also opened a Google form for other dog owners whose pets have been diagnosed with cancer.
Sam Altman, after meeting Conyngham, suggested the approach “should be a company” — signalling that OpenAI sees a commercial pathway in AI-assisted personalised medicine research.
Conyngham didn’t replace scientists — he orchestrated them. The vaccine was manufactured by Professor Thordarson. The genome sequenced by Professor Smith. AI gave one determined person the ability to build a research network without being part of one.
Conyngham moved fast partly because this was a dog. The same approach for a human patient faces significantly higher legal, ethical, and institutional barriers. The distance between “possible” and “permitted” is where the next chapter of this story lives.
True cost: $20,000–$50,000. Required months of time and world-class researcher connections. The ceiling on what one determined person can attempt has risen. The floor has not yet dropped.
Leave you with this
If AI can help one determined non-expert design a personalised cancer treatment — what else is it already doing that we haven’t heard about yet?
Three researchers weighed in. Their answers are not what you might expect.
“AI transforms a needle-in-a-haystack search into a data-driven selection process, drastically shortening the timeframe between diagnosis and vaccine construction.”
Prof. Patrick Tang Ming-kuen — Chinese University of Hong Kong“We don’t know enough about the vaccine to understand how much AI helped in its development — or if the vaccine worked the way it was designed.”
Nick Semenkovich — Medical College of Wisconsin“It was really driven by his determination to help his dog. This was not a clinical trial by any means. It’s not that AI cured cancer.”
Prof. Martin Smith — UNSW, who sequenced Rosie’s genome