Overview
In March 2026, a story originating from Sydney, Australia captured global attention across scientific, technology, and mainstream media circles. Paul Conyngham, co-founder of Core Intelligence Technologies and a former director of the Data Science and AI Association of Australia, used a combination of publicly available AI tools — ChatGPT, AlphaFold, and Grok — alongside genome sequencing to help design a personalized mRNA cancer vaccine for his rescue dog, Rosie. The dog had been diagnosed with aggressive mast cell cancer in 2024 and given only months to live after conventional chemotherapy and surgery failed to shrink her tumors. Following the vaccine’s administration in December 2025, Rosie’s tumor reportedly shrank by approximately 75% (The Decoder).
The story was amplified by OpenAI President Greg Brockman and DeepMind CEO Demis Hassabis, signaling that this was not merely a human-interest story but a signal event in the trajectory of AI-assisted biomedical research.
Related: Nvidia Bets $26 Billion on Open-Source AI to Fill the Gap OpenAI and Meta Left Behind
What Actually Happened: The Technical Breakdown
Understanding why this case matters requires a clear-eyed look at what Conyngham actually did, step by step.
Genome Sequencing as the Foundation
Conyngham had both Rosie’s healthy genome and tumor genome sequenced at the Ramaciotti Centre for Genomics at UNSW Sydney, at a cost of approximately $3,000. This produced gigabytes of raw genetic data — the kind of dataset that, until recently, required institutional infrastructure and specialist bioinformaticians to interpret (IBTimes UK).
ChatGPT as a Research Navigator
Conyngham used ChatGPT not as a passive information retrieval tool but as an active research partner — guiding the analytical steps, interpreting genomic data, and helping identify mutations potentially linked to the cancer. He described the process as “AI within AI within AI,” with ChatGPT orchestrating the broader research workflow (ABC Listen).
Related: ChatGPT’s Slipping Dominance: A Comprehensive Market Analysis of the AI Chatbot Landscape in 2026
AlphaFold for Protein Structure Prediction
Once mutations were identified, Conyngham used Google DeepMind’s AlphaFold to predict the 3D structures of the proteins encoded by those mutated genes. This allowed him to identify which protein structures might serve as viable neoantigens — targets for an immune response — and to match them with potential therapeutic agents (India Today).
Grok for Vaccine Design
The final vaccine design was produced using xAI’s Grok model. The resulting mRNA vaccine was manufactured with the help of researcher Pall Thordarson at UNSW and colleagues at the University of Queensland. The vaccine was administered to Rosie starting in December 2025, with booster shots following (The Decoder).
Ethics Approval: The Regulatory Bottleneck
Conyngham spent three months preparing a 100-page ethics approval document — dedicating two hours every night — to obtain permission to run a drug trial on Rosie. This regulatory friction is a critical data point: the science moved faster than the institutional frameworks designed to govern it (India Today).
Why This Case Matters
Democratization of Biomedical Research
The most significant implication of this case is not the specific outcome for one dog, but what it reveals about the changing accessibility of advanced biomedical research. As Saul Garcia Huertes noted on LinkedIn, “A few years ago this kind of exploration would have required large labs, big budgets and years of work. Today someone with enough curiosity can at least start exploring the same frontier.” (LinkedIn)
Conyngham had no formal background in biology or oncology. He is a machine learning consultant. The fact that he could navigate genomic data, protein structure prediction, and mRNA vaccine design — even with institutional collaboration — represents a qualitative shift in who can participate in biomedical discovery.
AI as a Force Multiplier for Non-Specialists
This case demonstrates that AI tools can function as a force multiplier for domain-adjacent experts. Conyngham’s ML background gave him the technical literacy to use these tools effectively, but the tools themselves lowered the barrier to entry for the underlying biology. This has direct implications for how we think about citizen science, patient advocacy, and the future of personalized medicine.
Related: Type-Safe LLM Pipelines With Outlines and Pydantic: Stop Parsing JSON With Regex
A PMC-published review on AI-powered personalized cancer vaccine design confirms this trajectory: “AI can assist in feature extraction and model training to predict patient-specific cancer antigens. Through sophisticated algorithms, AI may optimize and refine these antigens, guide vaccine formulation, and support clinical trial design, potentially enabling more personalized vaccination strategies.” (PMC)
What Changed: The Shift in AI Capability
From Institutional to Individual
Prior to tools like AlphaFold (publicly released by DeepMind in 2021) and large language models like ChatGPT and Grok, the workflow Conyngham executed would have required:
| Capability | Pre-AI Requirement | AI-Enabled Approach |
|---|---|---|
| Genomic data interpretation | Specialist bioinformatician | ChatGPT-guided analysis |
| Protein structure prediction | Months of computational work | AlphaFold (hours) |
| Vaccine sequence design | Dedicated research team | Grok model |
| Literature synthesis | Extensive manual review | LLM-assisted synthesis |
This table illustrates a compression of both time and expertise requirements that is genuinely unprecedented.
AlphaFold’s Role Is Structural, Not Incidental
AlphaFold’s contribution here is worth isolating. The ability to predict 3D protein structures from amino acid sequences — a problem that stumped structural biology for decades — is what made neoantigen identification tractable for a non-specialist. Without AlphaFold, Conyngham would have had mutation data but no reliable way to assess which mutations produced structurally distinct proteins that the immune system could target (LinkedIn - Sebastian Kmiecik).
Competitive Context: The AI Drug Discovery Landscape
This case does not exist in a vacuum. It sits within a rapidly expanding market for AI-powered drug discovery.
Market Scale and Growth
The global AI in drug discovery market is estimated to reach USD $3 billion by 2024, growing at a CAGR of approximately 40% due to increasing investments and technological advancements. A separate forecast projects the AI-powered drug discovery software market growing at a CAGR of 6.1% from 2026 to 2033 (LinkedIn Market Overview). Oncology holds the largest share of this market in 2025, driven by the complexity and heterogeneity of cancer as a disease (Maximize Market Research).
Key Players in the Institutional Space
| Platform | Focus Area | Key Technology |
|---|---|---|
| Insilico Medicine | End-to-end drug discovery | Pharma.AI suite |
| Atomwise | Biotech startups, rare diseases | AtomNet binding prediction |
| BenevolentAI | Drug repurposing, rare diseases | Massive dataset analysis |
| Recursion Pharmaceuticals | Rare diseases, oncology | LOWE LLM for dataset querying |
| Unlearn.AI | Clinical trial optimization | Digital twin technology |
What Conyngham did was effectively replicate — at a fraction of the cost and with publicly available tools — a workflow that these institutional platforms charge custom enterprise pricing to deliver. This is a competitive signal that should not be ignored by incumbents in the space.
Critical Assessment: Where the Skeptics Have a Point
The story has attracted significant skepticism from credentialed researchers, and their objections deserve serious consideration.
The Confounding Variable Problem
Egan Peltan, a Stanford-trained PhD in chemical biology and biotech co-founder, argues that Rosie was receiving conventional immunotherapy concurrently with the mRNA vaccine. This means there is no controlled way to attribute the tumor shrinkage to the vaccine specifically. Peltan also estimates the true cost of the treatment at $20,000–$50,000, significantly higher than the $3,000 sequencing cost that dominated media coverage (The Decoder).
Safety and Efficacy Remain Unproven
Patrick Heizer, a researcher in cell and gene therapy, raises a more fundamental concern: proteins in the body often share structural similarities, meaning a therapy targeting a tumor protein could inadvertently affect similar proteins in healthy organs such as the heart. He also notes that results in animals do not translate directly to humans due to protein differences across species (The Decoder).
The PMC review on personalized cancer vaccines corroborates this: “AI-driven models, despite their predictive power, are often constrained by the availability and quality of data, and their predictions may not fully account for the dynamic nature of the immune system and tumor microenvironment.” (PMC)
The Phase 3 Problem
Personalized mRNA cancer vaccines have been in development for years with no clear success in large-scale trials. Peltan’s core argument is that the field needs Phase 3 results, not anecdotes, before drawing regulatory or clinical conclusions. This is a legitimate scientific standard that the Rosie case, however compelling, cannot meet.
Practical Implications for AI Tool Users
What This Means for ML and Data Practitioners
For professionals with backgrounds in machine learning, data science, or software engineering, this case establishes a concrete proof of concept: domain-adjacent AI literacy can be applied to biomedical problems that were previously gated behind years of specialist training. The tools are available. The sequencing infrastructure is accessible. The limiting factor is now primarily regulatory and methodological, not technical.
The Regulatory Gap Is the Real Bottleneck
Conyngham’s three-month ethics approval process for a single dog trial is a telling data point. The scientific workflow took months; the regulatory workflow took the same amount of time. As AI compresses the research timeline further, the regulatory infrastructure will increasingly become the rate-limiting step. This has implications for policy, for biotech startups, and for patient advocacy organizations pushing for faster access to experimental treatments.
Collaboration Remains Essential
It is important to note that Conyngham did not work alone. He collaborated with researchers at UNSW and the University of Queensland. The AI tools enabled him to arrive at the collaboration with a credible hypothesis and a structured research plan — but the actual vaccine manufacturing and administration required institutional expertise. This is the realistic model for near-term AI-assisted biomedical research: AI as the entry point, human expertise as the execution layer.
Where This Fits in the Market
This case represents a convergence of three trends that have been building independently:
- The commoditization of genomic sequencing — $3,000 for whole-genome sequencing of a tumor is accessible to motivated individuals, not just institutions.
- The public availability of frontier AI tools — AlphaFold is open-source; ChatGPT and Grok are consumer products. The barrier to access is a subscription fee, not a research grant.
- The maturation of mRNA vaccine technology — COVID-19 accelerated the manufacturing and regulatory infrastructure for mRNA therapeutics, making bespoke vaccine production more feasible.
The intersection of these three trends is what made Conyngham’s project possible in 2025–2026 rather than 2015. The AI drug discovery market’s projected growth at 40% CAGR reflects institutional recognition of this convergence. What the Rosie case adds is evidence that the convergence is already producing results outside institutional settings.
Conclusion
The Rosie case is neither a cure nor a clinical trial. It is a proof of concept with significant methodological limitations. However, dismissing it as mere anecdote misses the structural significance of what occurred: a non-biologist, using publicly available AI tools and a $3,000 sequencing investment, produced a personalized mRNA cancer vaccine that was administered in a university-supervised trial and yielded measurable tumor reduction. The real takeaway, as The Decoder noted, is that “AI enabled a medical layperson to get this far in the first place.” That is the signal. The noise is the debate over whether the vaccine specifically caused the shrinkage.
For AI practitioners, biotech investors, and policy makers, the question is no longer whether AI can assist in personalized medicine — it demonstrably can. The question is how to build the regulatory, ethical, and institutional frameworks that allow this capability to be deployed safely and at scale.
Next Step
Use these pages to keep the decision moving:
- More in Coding — Keep researching the same category instead of stopping at one article.
- Open comparisons — Jump into direct matchups and trade-off pages.
- Open tool guides — Use the canonical decision pages for fit, pricing context, and alternatives in one place.