When Does an AI Scribe Become a Medical Device in Britain?

In Great Britain, an AI scribe becomes a medical device when its manufacturer’s intended purpose meets the medical-device definition, not simply because the software is used in a clinic. Our analysis is that buyers should separate that classification decision from the continuing work of deploying, checking and updating the product safely.

Start with the purpose the supplier actually claims

The MHRA says a product’s intended purpose is defined by its instructions, labelling and promotional materials; use in a medical environment alone does not make software a medical device (The ambient-technology guidance). Our recommendation is to put that purpose in the purchasing record alongside the functions being bought.

The MHRA publishes separate guidance on crafting an intended purpose for software as a medical device (Its intended-purpose publication). Buyers could ask the supplier to explain how each proposed function fits that statement. The practical issue is the product’s claimed role, not whether a demonstration sounds conversational.

A disclaimer cannot erase a medical function

The MHRA’s borderline-products guidance bases device decisions on intended purpose and mode of action (The regulator’s borderline guidance). In our assessment, procurement teams should compare the sales demonstration with the written purpose and instructions, rather than classify the system by its marketing label.

NHS England asks organisations to review functionality changes that may alter an ambient product’s regulatory status (NHS England’s regulatory note). We would therefore require a documented change-review process. A purchase decision should not be treated as a permanent answer for every later feature.

Clinical review remains part of the workflow

NHS England’s professional guidance tells clinicians to check whether their organisation has approved the product and what local rules govern its use (The clinician guidance). Our recommendation is to demonstrate the whole workflow, including correction and approval of a draft record.

The MHRA says clinicians remain responsible for reviewing and verifying AI-generated transcripts, summaries and other outputs before they are used in patient care (The regulator’s July announcement). For buyers, that makes review effort a diligence question alongside the time spent generating text. Neither a polished demo nor a regulatory label establishes local productivity gains.

Data handling is a separate assurance exercise

NHS information-governance guidance recommends updating privacy notices and explaining how people can request that ambient scribes are not used (The information-governance guidance). We would map recording, processing, review and storage before setting an implementation timetable.

NHS England describes ambient scribing products as converting speech into draft summaries for review and validation (The adoption overview). Our analytical recommendation is to measure the approved record, not just the generated draft. A pilot could examine correction work, patient communication and integration with the clinical record, without assuming that results elsewhere will transfer unchanged.

The registry does not replace local responsibility

The supplier registry states that all assurance and decision-making sits with the local NHS organisation, not NHS England (The self-certified registry). NHS England’s deployment guidance says a pilot does not exempt a supplier from compliance (The wider deployment guidance). A trial should therefore have defined responsibilities before it starts.

Our conclusion is to buy a specified workflow with an accountable owner and a route for reviewing updates. The analysis concerns Great Britain’s device guidance and NHS England’s deployment advice, not a universal regulatory rule. It offers procurement questions, not a finding that any named supplier is compliant, clinically superior or suitable for a particular patient.

Related reading: AI-device change planning, clinical innovation, maternal AI research, medical-device reporting, administrative healthcare AI. These links provide wider business context, not independent evidence for the sources examined here.