ChatGPT: No Miracle for Dog Cancer

The compelling narrative of a gravely ill dog, a determined owner, and the supposed intervention of chatbots captivated many, though the underlying scientific reality proved considerably more intricate.

When an Australian technology entrepreneur, lacking any background in biology or medicine, claimed that ChatGPT played a pivotal role in saving his dog from cancer, the story rapidly gained traction. This account offered the kind of validation eagerly sought by major tech firms: apparent proof that artificial intelligence could revolutionize medicine and tackle some of its most formidable diseases. However, the truth, as often happens, presented a more complex picture.

The version of the story widely circulated online, initially reported by The Australian, was relatively straightforward. In 2024, Sydney-based Paul Conyngham discovered his dog, Rosie, had cancer. Despite chemotherapy slowing the disease, it failed to reduce the tumors. After veterinarians advised that "nothing could be done" for his Staffordshire bull terrier-shar pei mix, Conyngham stated, "I took it upon myself to find a cure."

Conyngham recounted using ChatGPT to brainstorm treatment concepts. The chatbot suggested immunotherapy as a potential avenue and directed him to specialists at the University of New South Wales (UNSW), who subsequently performed genetic profiling of Rosie’s cancer. He then leveraged ChatGPT and Google’s protein structure AI model, AlphaFold, to help interpret the results. Collaborating with UNSW professor Pall Thordarson, Conyngham pursued a personalized mRNA vaccine specifically designed to target Rosie’s tumor mutations. Thordarson informed The Australian that he believes this marks the first instance of such a treatment being developed for a canine.

Just weeks after Rosie received her first injection last December, Conyngham reported that her tumors had diminished, and her condition had improved, with her even chasing rabbits in the park. Nevertheless, the tumors had not entirely vanished, and one showed no response. "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," Conyngham told The Australian.

This crucial nuance was largely lost as the story spread. Newsweek published the headline, "Owner With No Medical Background Invents Cure for Dog’s Terminal Cancer," while the New York Post declared, "Tech pro saves his dying dog by using ChatGPT to code a custom cancer vaccine." On social media, numerous accounts amplified Rosie’s case as a "cure" and a harbinger of a new era in personalized medicine. Some individuals, notably OpenAI president and cofounder Greg Brockman, arguably should have exercised more caution, while others, like Google DeepMind CEO Demis Hassabis, shared it without excessive hype. Elon Musk also weighed in, eager to highlight that xAI’s Grok had a role—a detail often omitted from initial coverage.

Furthermore, the narrative attributes far too much credit to AI. Not only was Rosie not cured of cancer, but it remains uncertain whether the mRNA vaccine was solely responsible for her improvement. The personalized treatment was administered concurrently with another form of immunotherapy, a checkpoint inhibitor, designed to help the immune system target tumors. This co-administration makes it challenging to ascertain the vaccine's individual efficacy. Martin Smith, one of the involved scientists, stated that the team is conducting tests to assess the immune response.

ChatGPT did not design or create Rosie’s treatment; human researchers were responsible for this endeavor.

Nor was the vaccine itself generated by a chatbot. ChatGPT did not design or create Rosie’s treatment; human researchers did. At most, the chatbot functioned as a research assistant, helping Conyngham navigate medical literature—an impressive capability, but a far cry from the implied breakthrough.

Reports are also ambiguous regarding AlphaFold’s specific contribution. David Ascher, a professor and director of biotechnology programs at the University of Queensland in Australia, informed The Verge that the model "could contribute structural hypotheses about proteins, but it is not a turnkey cancer-vaccine design system." He also noted that official guidance warns AlphaFold is not validated for predicting the effects of certain mutations and does not model "several biologically important contexts."

Grok’s involvement is even more difficult to precisely define. Conyngham posted on X that "the final vaccine construct for Rosie was designed by Grok," yet the practical implications of this statement and the inputs provided to the model are unclear. Ascher suggested that Grok would realistically fall into a similar category as ChatGPT: a tool that "could help with literature search, summarising papers, translating jargon, suggesting workflows, drafting code or documents, and helping a user think through options." While useful, this role is vastly different from "designing a cancer vaccine."

The "AI made this" framing entirely overlooks the extensive human effort, without which "AI’s output would have remained just text on a screen."

Overall, Ascher characterized Rosie’s case as "better seen as an unusual, highly specific proof of possibility than a template ordinary people can readily reproduce." He emphasized that it necessitated "substantial" expert labor, "not just a chatbot and a few prompts."

This distinction holds particular importance in medicine, where success hinges not merely on generating plausible information but on the expert, physical work involved in producing, testing, and delivering actual treatment. Alvin Chan, an assistant professor at Nanyang Technological University in Singapore who develops AI for biomedical and drug discoveries, told The Verge that the "AI made this" framing disregards this massive human effort. Without it, he contends, "the AI’s output would have remained just text on a screen." In Rosie’s situation, AI is more accurately understood as a tool for outlining a blueprint rather than the creator of the treatment itself.

The entire episode carries a subtle, persistent suggestion of a public relations maneuver. Bold assertions built upon questionable foundations and employing vague methodologies often align conveniently with the world of tech fundraising. mRNA vaccines, much like the broader promise of personalized medicine, remain largely unproven as cancer treatments in humans, let alone dogs. While the case itself may be genuine, it appears overly neat and conveniently glosses over the tens of thousands of dollars and extensive expertise required to transform the initial concept into a viable treatment.

I reached out to Conyngham via X requesting an interview but have not received a response. His profile states "Ending Cancer for Dogs" and directs to a Google form outlining his "dream to make this process something everyone could have access to." The form inquires whether one's dog has cancer, whether one is a researcher or scientist interested in involvement, and whether one is an investor.

It would be a mistake, however, to entirely dismiss Rosie’s story as meaningless. While AI may not be replacing laboratory work anytime soon, it is undoubtedly enhancing the accessibility of scientific information for ordinary individuals. Yet, this accessibility to information is not synonymous with accessible care. Few patients—or pet owners—possess ready access to the world-class experts, specialized equipment, and substantial financial resources necessary to translate that information into tangible treatment.