Study Finds 63 Percent of Veterinary AI Vendors Disclose No Validation Data as Industry Transparency Push Gains Momentum
A systematic audit of 71 commercial veterinary AI products published in Frontiers in Veterinary Science found a mean transparency score of just 6.4 percent, prompting at least one company to expand its public performance dashboard in response.
A Market With Almost No Public Accountability
The commercial veterinary AI market in North America encompasses at least 71 products, yet the vast majority of vendors provide no public evidence that their tools work as advertised. A systematic audit published in Frontiers in Veterinary Science on March 5, 2026 evaluated these products using a 25-point framework adapted from FDA and Good Machine Learning Practice guidelines. The results paint a stark picture: the mean transparency score across all products was 6.4 percent, and 63.3 percent of vendors disclosed zero performance metrics.
The study, authored by David Brundage at the University of Wisconsin-Madison School of Veterinary Medicine, sorted the products into three categories: generative and ambient tools (47 products), diagnostic imaging (19 products), and specialized tools (5 products). Diagnostic imaging vendors scored notably higher, with a mean risk-weighted transparency score of 13.1 percent and 36.8 percent of vendors providing some validation evidence. Generative and ambient tools fared far worse, achieving a mean score of just 1.8 percent, with only 2.1 percent of vendors in that category disclosing validation data.
The Transparency Gap
Brundage characterizes the core finding as a “Transparency Gap” — the disparity between the clinical capabilities that vendors market to veterinarians and the publicly available evidence backing those claims. Unlike the human medical device market, where the FDA requires premarket screening of AI-enabled devices, no equivalent federal approval requirement exists for veterinary AI tools in North America. This regulatory absence creates what the study describes as an accountability vacuum, leaving veterinarians to assume legal and ethical responsibility for validating technologies whose performance data they cannot access.
A particularly concerning finding is the near-universal failure to report training data demographics. Only 1.4 percent of vendors disclosed information about the species, breed, age, or sex composition of their training datasets, making independent assessment of algorithmic bias effectively impossible.
Professional Bodies Sound the Alarm
The audit builds on mounting institutional concern. In a joint position statement published in the Journal of the American Veterinary Medical Association, the American College of Veterinary Radiology and the European College of Veterinary Diagnostic Imaging concluded that no commercially available AI product for veterinary diagnostic imaging met their required standards for transparency, validation, or safety. The statement, authored by Ryan Appleby and colleagues, called for rigorous peer-reviewed research, unbiased third-party evaluations, and the maintenance of a “veterinarian in the loop, preferably a board-certified radiologist” to oversee AI outputs.
The ACVR and ECVDI also recommended that veterinarians disclose AI usage to pet owners and offer alternative diagnostic options, and urged regulatory bodies to establish guidelines to prevent misuse.
One Company Responds
Vetology Innovations, a veterinary radiology AI provider, has moved to address the gap. On April 1, 2026, the company expanded its public validation dashboard from 4 to 11 metrics per classifier, covering 89 validated classifiers across canine and feline imaging. The dashboard now reports sensitivity, specificity, positive predictive value, negative predictive value, area under the curve, F1 score, accuracy, prevalence, confidence intervals, and radiologist agreement rate. All classifiers were revalidated with confusion matrices as of February 2026, drawing on a dataset of 300,000 multi-image patient cases.
The update also introduced six new classifiers, including models for obscuring pleural effusion, esophageal enlargement, and intervertebral disc disease, along with 31 retrained and revalidated models. Vetology’s Chief Technical Officer, Cory Clemmons, stated that metrics like “PPV, confidence intervals, specificity — that’s what lets a veterinarian decide how much weight to put on what the model is telling them.”
What Remains Unresolved
Vetology’s response remains an outlier. The Frontiers audit found that across the broader market, the overwhelming majority of products — particularly the fast-growing category of generative and ambient AI tools used for medical records and clinical summaries — operate with effectively no public accountability. Whether the convergence of the Frontiers audit, the ACVR/ECVDI position statement, and Vetology’s transparency push will catalyze wider industry disclosure or prompt regulatory action remains an open question.