Artificial intelligence is already embedded in daily life - from facial recognition unlocking your smartphone, to the satnav directions, social media and search engine summaries, there is progressively less human intent and more algorithm-driven determination of the information we consume. When it comes to work, it's no surprise that AI use is creeping into daily use - systemically or otherwise. Used well, it can improve efficiency. Used covertly, it creates a far more serious problem: shadow AI.
Shadow AI is the use of AI outside approved systems, policies, governance and audit trails. A team member pastes a case history into a public chatbot to draft a referral. A clinician uses an unapproved tool to summarise records. A foundation model is used to generate a differential list.
Why Are Workers Resorting to Using AI in the Shadows?
A recent study from PagerDuty found that among office employees, 66% of office employees were using AI covertly1. About a third cited restrictive company policies or concern about the opinion of their co-workers as reasons for hiding their use of AI, while 29% expressed uncertainty around company policies.
Whatever the reason for 'shadow AI' use, the vast majority of workers believed the leadership of their company trusted them to make informed decisions about which AI tools work best for their daily work - up to 94% in the UK. For a sector with regulated professionals, where the only liability currently lies with the individual, it's perhaps a natural assumption that use of AI by clinicians will be responsible, considered and informed.
The Missed Potential and Risks of Covert AI Use
AI in veterinary practice suffers from a lack of transparency, training data and informed use. As a result, shadow AI is the worst possible AI pattern of use for veterinary medicine - not only because the organisation cannot see, assess or control what is happening, but it also presents a huge missed opportunity in sharing best practice and optimising development and use.
-
Intellectual property and confidential information may be exposed. Veterinary records typically contain client information and commercially sensitive pricing. Freemium AI services, including mainstream models, handle content very differently from enterprise systems, and depend on the provider, product and user settings. Once sensitive material has been entered into an unapproved external tool, the organisation loses sight and control over it.
-
Quality controls are bypassed. Generative AI can produce fluent, convincing content that is incomplete, inaccurate or simply wrong. Mainstream models are trained on such a broad corpora of texts that accurate veterinary information is a minority. In a clinical setting, that might mean an inaccurate summary, a misleading client communication, an omitted differential diagnosis or an unsafe clinical recommendation.
-
Shadow AI opens individuals up to regulatory and professional exposure. A clinician needs to be transparent about how information is shared and decisions are reached. If hidden AI use is uncovered through retrospective investigation of errors or complaints, “We did not know staff were using it” is not an adequate governance position for practices.
-
Shadow AI prevents organisational learning. When use is hidden, no one can measure where AI is helping, where it is failing, what data is at risk, or where staff need better tools and training to optimise use. The business loses the opportunity to assess which applications are generating positive outcomes, such as improving workflows and supporting good decision making and patient care. Instead, AI use is a variable, fragmented, layer of practice dependent on individual confidence and preference, rather than robust, governed application.
Other sectors have already seen the warning signs. High-profile data-leak concerns led major technology companies to restrict employee use of public generative AI tools2, while UK government guidance explicitly warns staff not to place sensitive information into public tools3. Veterinary practices should avoid the assumption that staff are not using AI or instigate a blanket ban that may drive use further underground, but rather open discussion and develop governance around approved products, data security, clinical safety and human oversight.
Bringing AI Use into the Light
Enabling safe use of AI requires a clear, practical AI policy: define which tools are approved; specify data controls; require an appropriate level of human review of all clinical and client-facing outputs; make disclosure routine rather than punitive; and provide a safe route for staff to propose new use cases.
The goal is not to slow innovation, but rather to ensure that innovation remains visible, accountable and contributes to shared learning of best practice and optimal use. Shadow AI thrives where there is uncertainty. Clear governance makes it safe to speak up, test responsibly, learn and improve together.
At Vet Validaite, our aim is to boost the safe, confident adoption of AI in animal health by providing independent review of AI tools against best practice standards. Sign up to our newsletter to be the first to hear about AI tool validations - coming later this year! Plus, lots of useful information about safe use of AI for veterinary teams.