AI Is Curing Cancer and No One Is Talking About It: Why Personalized Vaccines Change Medicine

A guy in Sydney with zero medical training used AI to help design a personalized mRNA cancer vaccine for his dying dog. The tumors shrank 50 to 75%. Leaders at OpenAI, DeepMind, and xAI all shared the story. Most headlines still got the important part wrong.

Paul Cunningham is an AI consultant in Sydney. Seventeen years in machine learning. No biology degree. No medicine background. In 2019 he adopted Rosie, a Staffordshire bull terrier–Shar-Pei mix. In 2024 she was diagnosed with aggressive mast cell cancer—the most common skin cancer in dogs. The vet gave her one to six months. Tennis-ball-sized tumors. One massive growth on a hind leg. Most people would make her comfortable and wait. Paul went to ChatGPT.

What he actually did

Step one: he paid about A$3,000 to sequence Rosie's genome at UNSW Sydney's Center for Genomics. Healthy DNA and tumor DNA. Comparing the two shows the mutations driving the cancer. Those mutations create neoantigens—protein markers unique to the cancer cells. A fingerprint that says "this is the bad tissue."

Step two: he used ChatGPT to navigate biomedical literature he had never studied. It helped him ask better questions, build a data analysis plan, process genetic data, spot which mutations mattered, brainstorm treatments, and draft a half-page sequence—an early blueprint for an mRNA vaccine. A non-expert walked in with genome files and left with a working design sketch in hours. That used to take a decade of training just to ask the right questions.

Step three: he used Google DeepMind's AlphaFold to model the 3D structure of the mutated protein driving Rosie's cancer—the c-Kit protein. Shape matters in drug design. Shape decides how a protein works, what it binds, and where you can hit it. AlphaFold showed which parts of mutated c-Kit to target.

Step four: according to Paul, xAI's Grok designed the final vaccine construct. Elon Musk called out headlines that credited only ChatGPT when Paul himself said Grok built the final mRNA construct. OpenAI, DeepMind, and xAI each contributed a different piece. Greg Brockman, Demis Hassabis, and Musk all pointed at the same dog. That almost never happens across rival labs.

Then the humans and the red tape

Paul did not DIY the syringe. He worked with Professor Paul Thordarson, director of the UNSW RNA Institute—Australia's first dedicated RNA research center—with 190-plus publications and more than 12,000 citations. Lipid nanoparticles, the delivery bubble for mRNA, are his lane. He also worked with Professor Rachel Allavena at the University of Queensland's School of Veterinary Science, who leads canine cancer immunotherapy trials and co-founded Animal Immuno Solutions, plus Dr. Martin Smith at the Ramaciotti Centre on genomics.

Ethics approval took three months and a 100-page document. Paul said the red tape was harder than creating the vaccine. AI-assisted design of a personalized cancer vaccine moved faster than permission to use it. The tech is racing. The paperwork is not.

Rosie got her first dose in December 2025, then boosters in January and February 2026. The hind-leg tumor shrank by about half. Overall tumor reduction landed near 75%. She went from barely moving to jumping fences after rabbits.

The caveats that matter

She was also on conventional immunotherapy. There is no clean way to prove how much shrinkage came from the mRNA vaccine versus the standard treatment. N equals one. No control group. That is an anecdote with dramatic timing—not a peer-reviewed trial.

The "$3,000 cure" headlines were wrong. That figure was genome sequencing alone. Factor in mRNA production at UNSW's RNA Institute, donated professor time, ethics work, and vet administration, and serious estimates land roughly $20,000 to $100,000. Stanford PhD and biotech co-founder Egan Pelton puts this case nearer $20,000 to $50,000. Cancerhealth.com has floated ~$100,000 per person at current scale. One tumor did not respond at all. Paul is building a second vaccine aimed at different antigens. He is clear: this is not a declared cure. It bought Rosie time and quality of life.

Skeptics have fair points. Pelton argued much of the workflow could happen without ChatGPT—literature review, sequencing plans, and protein modeling are routine for experts. Mast cell tumors can rarely regress on their own, though vet literature calls that rare in adult dogs; untreated high-grade cases often have median survival under four months. Patrick Heiser, a cell and gene therapy researcher, flags off-target risk: tumor proteins can resemble healthy ones, so a vaccine could hit the wrong tissue. That safety worry is why human trials take years. Still, two senior professors put their names on the project. That is not a random influencer stunt.

Why this is bigger than one dog

Three stacks collided. First, AI: ChatGPT for research navigation, AlphaFold for structure, Grok for vaccine construct. Second, mRNA platforms built at scale during COVID—Moderna, BioNTech, and labs worldwide can now spin custom sequences fast. Third, cheap genomics. Rosie's tumor genome cost about A$3,000. The first human genome took 13 years and $3 billion ending in 2003. Twenty-three years later, a dog tumor sequence costs roughly a laptop.

Human trials are already past the "maybe someday" stage. KEYNOTE-942 (Moderna + Merck) tests a personalized mRNA vaccine—intismeran autogene—with Keytruda after melanoma surgery. Phase 2b five-year data showed about a 49% cut in recurrence or death risk versus Keytruda alone. Phase 3 was fully enrolled in early 2026. BioNTech and Genentech reported a pancreatic cancer Phase 1 where six of eight vaccine responders were cancer-free at three years. Kidney cancer, glioblastoma, and even dog glioblastoma trials are underway. Personalized mRNA cancer vaccines are in late-stage human work. Rosie's story adds the AI compression layer: shorter timelines and a lower expertise barrier to design.

Speed is the signal

Think self-driving cars five or six years ago. People argued about edge cases, liability, and whether the tech was real while the stack kept improving. The same fight is happening over personalized medicine: Did the vaccine work? Can it scale? Is regulation ready? Cost? Safety? Meanwhile sequencing, AI analysis, mRNA design, and manufacturing keep getting faster. Paul went from genome data to first dose in about two months—plus three months of paperwork. In an optimized setup, that design-to-dose loop could shrink further. The bottleneck is shifting from science to bureaucracy.

Regulatory systems were built for blockbuster drugs that treat millions and need decade-long trials. Personalized therapies are one-patient products. That mismatch is a hard policy problem, and AI is forcing the conversation early.

The honest frame is not "ChatGPT cured cancer" or "AI did nothing." Three specialized systems plus a determined human orchestrator got a non-biologist from raw data to a therapeutic candidate and university partners in months. Paul had the motive, the ML experience, and the grit to push ethics boards and campuses. The models did not file the 100-page form. Bridging expertise and action is the breakthrough—medicine first, then every field where a PhD used to be the ticket just to ask the right question.

Five years from now this may look like the first SpaceX landing: messy, incomplete, unmistakably early. Costs may stay high. One case may not scale. Or the pattern of every exponential stack may repeat—scrappy proof of concept, real results hard to dismiss, cost curves falling, capability rising, multiple actors converging. What Paul did for Rosie, someone will try for a parent or a child. The open question is how fast institutions catch the tech.

Zoom out past the hype and the doom. A man in Sydney used three AI systems to help design a cancer vaccine for his dog, and the tumors shrank. That happened in the real world. Pay attention.

Check the video here.