VTA Agenda
Responsible AI in Veterinary Medicine
Veterinary AI should be evaluated according to what it does, the risks it introduces and the evidence supporting its use.
The problem
AI can assist with clinical records, client communication, information retrieval, diagnostic interpretation and operational tasks. But the term "AI" covers systems with very different purposes and capabilities.
A tool that schedules appointments does not present the same clinical risks as one interpreting diagnostic images or recommending treatment.
An AI system that drafts a note also differs from an agent authorised to change a patient's record or initiate a clinical workflow.
Veterinary professionals need to understand these distinctions.
They also need reliable information about accuracy, limitations, oversight, data use and the responsibilities of suppliers.
Marketing claims alone cannot establish that a product is safe or effective for its intended use.
What needs to change
Risk proportionate to intended use
Expectations for testing, monitoring and oversight should reflect what a system is intended to do and the possible consequences of error. Administrative automation, clinical documentation, decision support and autonomous actions should not automatically be treated as equivalent.
Evidence of performance
Where AI suppliers make claims about accuracy or clinical performance, those claims should be supported by appropriate evaluation. Testing should reflect the intended veterinary use, including relevant species, clinical settings and patient populations.
Clinical oversight
Veterinary professionals should understand where AI is being used, what outputs require review and what actions a system is permitted to perform. Clinical accountability must remain clear.
Transparency about capabilities and limitations
Suppliers should explain a product's intended use, important limitations, known failure modes and the role expected of the veterinary professional. The use of AI should not be concealed behind claims that imply a level of reliability, validation or autonomy that has not been demonstrated.
Responsible data use
Clinical records and related information may be used for different purposes, including providing a service, evaluating performance or developing models. Suppliers should be clear about those purposes, the permissions and legal bases involved, and the safeguards protecting the information.
Monitoring after deployment
Performance in a controlled test does not necessarily establish performance in routine clinical practice. The VTA should consider expectations for incident reporting, error correction, monitoring and communicating material changes to AI systems.
Questions for members
- How should veterinary AI applications be categorised by risk?
- What evidence should be expected before a supplier makes a clinical performance claim?
- What must veterinary professionals be able to review, correct or override?
- How should suppliers communicate significant errors and limitations?
- What constitutes meaningful transparency about training data, evaluation and use of clinical information?
- How should AI systems taking actions through practice software be authenticated, permissioned and audited?
- Which existing research and professional guidance should inform a common industry approach?
What we want to work towards
A practical, risk-based framework that helps technology suppliers develop responsible products and veterinary professionals evaluate them.
This work should draw on veterinary research, clinical expertise and existing regulatory and professional guidance.
The VTA does not intend to replace regulators, professional bodies or independent scientific evaluation.
Its contribution is to bring developers and veterinary users together around practical expectations that can be understood and applied.
Membership is open
Join the founding cohort of the Veterinary Technology Association and help set the agenda rather than react to it.