AI-Designed mRNA Vaccine Shrinks a Dog's Tumor 75%: What the Case Actually Shows
Agentic AIAn Australian AI consultant used a chain of AI tools to design a personalized mRNA cancer vaccine for his dog — and the tumor reportedly shrank by 75 percent. The case is striking as an example of agentic AI applied to a high-stakes real-world problem. The expert reactions are equally instructive about where the limits still are.
Context
Personalized cancer vaccines — treatments engineered to target the specific mutational profile of an individual tumor — represent one of the more promising directions in oncology. The challenge has historically been cost and speed: genome sequencing, protein structure analysis, and vaccine design require specialized expertise and laboratory infrastructure that puts the approach well outside reach for most patients and their animals.
The case of Paul Conyngham and his dog Rosie is notable because it demonstrates that a sufficiently capable AI operator can now assemble a functional pipeline for this process using commercially available tools — without institutional access to a research lab or oncology team. Whether the outcome is attributable to that pipeline is a separate, contested question.
The Detail
Conyngham, an AI consultant, began the project in November 2024 after Rosie was diagnosed with mast cell cancer. The workflow he assembled moved through four stages: consulting ChatGPT for initial treatment recommendations and approach guidance; commissioning genome sequencing of both Rosie’s healthy and tumor tissue at UNSW Sydney’s Ramaciotti Centre for Genomics at a cost of $3,000; using Google DeepMind’s AlphaFold for protein structure prediction to identify target proteins in the tumor; and using Grok AI for the final vaccine design. Treatment was administered by January 2025.
The reported outcome is a roughly 75 percent reduction in tumor size. Full remission has not been achieved.
Expert reactions split along two lines. Patrick Heizer, a cell and gene therapy researcher, offered a cautionary read: success in a non-controlled case does not demonstrate safety or efficacy. Fighting tumors in a laboratory context is the tractable part; the harder problems are proving that the intervention doesn’t accidentally target healthy tissue and establishing that the result is reproducible. Egan Peltan, a biotech co-founder, went further, arguing that AI’s role in the outcome was “overblown” — conventional immunotherapy was administered concurrently, making it impossible to isolate the vaccine’s contribution to the tumor reduction.
The So What
The honest read on this case is that it demonstrates a proof-of-concept for a new kind of agentic AI workflow — not a validated medical treatment. Conyngham assembled a pipeline that would have required a team of specialists and institutional infrastructure a few years ago, using tools that cost a few thousand dollars and are accessible to anyone with the technical fluency to chain them together. That’s a meaningful demonstration of what agentic AI makes possible at the frontier.
What it doesn’t demonstrate is that the pipeline works in the sense that matters for medicine: controlled, reproducible efficacy with known safety profiles. The experts who raised flags about healthy tissue targeting and confounded treatment attribution aren’t being dismissive — they’re describing the exact problems that have caused promising preclinical interventions to fail in trials for decades. The signal here is about capability and access, not clinical proof. Both things can be true simultaneously, and distinguishing them is the work.
Photo by Tima Miroshnichenko on Pexels
Content created with AI assistance and reviewed for accuracy.
Join the conversation
Stack Insiders is our free community for readers who want to go deeper — share resources, ask questions, and connect with others across every vertical we cover.
Join Stack Insiders →