AI in UK telehealth is a category that has shifted from speculative to deployment over twenty-four months. The technology has matured, the use cases have narrowed to the genuinely useful, and the regulatory line — what counts as a medical device under MHRA — has become clearer. This piece is the operator's view of where AI works, where it does not, and how to deploy it without crossing into medical-device territory by accident.
Where AI works in UK telehealth right now
Four realistic AI use-cases in 2026. Intake-flow optimisation — using AI to surface red flags from questionnaire responses for clinician attention. Post-consultation summarisation — drafting structured notes from synchronous calls for clinician review. Clinical coding automation — mapping diagnoses to SNOMED codes. Triage support — pre-sorting tickets and queries by likely category for human routing. None autonomously make clinical decisions. All add operational capacity without crossing regulatory lines.
Where AI does not work — autonomous prescribing, unsupervised triage
The use cases that look attractive but are not viable: autonomous prescribing decisions, unsupervised patient triage, AI-only diagnosis, automated dose adjustment without clinician sign-off. The reasons are regulatory and practical. Regulatorily, these likely cross into medical-device territory requiring MHRA certification. Practically, the failure modes are clinical not technical and the liability allocation is unresolved. Pursuing them is a strategic mistake regardless of model capability.
The MHRA medical-device line — what crosses it
AI-assisted clinical decision support tools must be certified as medical devices under UK MDR where they make clinical decisions. The boundary: tools that assist a clinician retain device-status exemption; tools that substitute for clinical judgement are software-as-a-medical-device (SaMD). The SaMD classification (Class I, IIa, IIb, III) drives the regulatory burden, the evidence requirements, and the post-market surveillance obligations. Get the classification wrong and the product launch becomes a regulatory remediation project.
Vendor due diligence on AI tooling
AI vendors selling into UK telehealth come in three flavours. Legitimate medical-device-certified products with appropriate clinical evidence and post-market surveillance. Clinical-decision-support tools positioned just below the medical-device line, sometimes credibly and sometimes not. General-purpose LLM wrappers marketed for healthcare, often without serious regulatory consideration. Diligence them all on: regulatory classification, evidence base, training-data provenance, UK GDPR posture, and clinical-governance review. The third category dominates marketing volume and is the highest-risk.
Data governance for AI in telehealth
Patient data flowing to AI tooling raises specific UK GDPR concerns. Special-category data processing under Article 9 must have a clear lawful condition documented before processing starts. Training-data use must be documented in the DPA with the vendor. Patient consent for AI-assisted processing must be informed, not buried in terms and conditions. Cross-border data flows to AI providers need adequacy decisions or appropriate safeguards. Done well, this is not a blocker — done casually, it is an ICO enforcement risk.
What to deploy this year — and what to wait on
Deploy now: intake red-flag detection with clinician review, post-consultation summarisation with clinician sign-off, triage pre-sorting with clear escalation routes, clinical coding suggestions with audit trail. Wait on: any tool that substitutes for clinician judgement, any tool that markets autonomy, any tool without a clear medical-device classification. The conservative deployment list is the one operators look back on without regret. PExpo's data architecture supports governed AI deployment via REST API and signed webhooks — see our integrations page.
AI-assisted clinical decision support tools must be certified as medical devices under UK MDR where they make clinical decisions. The SaMD classification drives the regulatory burden. Get it wrong and the product launch becomes a remediation project.
AI in regulated health works when it amplifies the clinician — not when it replaces them. The hard part is staying disciplined about the difference.
AI in UK telehealth in 2026 is genuinely useful, narrowly. The operators who deploy it well stay tightly within the assistive-tool boundary and treat the medical-device line as the firm constraint it is. The ones who chase autonomy and marketing claims discover the regulatory implications too late. Our integrations page covers the data architecture that makes governed AI deployment possible, and our brand model page shows the broader operational scope where AI fits as one component among many.
