MHRA's AI plan sets 3 dates. None bind you.

On 14 October 2026 the MHRA published its strategy to 2030, promising a "globally leading" approach to regulating AI as a medical device (MHRA Strategy 2030 MedTech explainer, GOV.UK, October 2026). In the press release and the explainer we read, ambient scribes are not mentioned, and neither are the practices that switch them on. Three dates in the past fortnight's announcements touch AI devices, and each is addressed to developers or to government, not to a practice director deciding whether to keep a scribe running next term.

If you direct a private psychology practice, the question worth answering is short: does anything announced this month change what you must do before, or while, an AI scribe is in use? On the evidence available today, no. The reasons are worth setting out, because two of the three dates deserve a place in your diary.

What the MHRA actually published on 14 October

For AI devices, the strategy commits the MHRA to an adaptive regulatory framework, a new Centre for AI and Digital Health to clarify expectations for developers, continuous monitoring of real-world performance, staged authorisation, expanded predetermined change control plans (pre-agreed model updates) and exploration of master files for the foundation models that sit under products (MHRA Strategy 2030 MedTech explainer, GOV.UK, October 2026). Chief executive Lawrence Tallon said AI is "transforming both regulated products and how organisations are run" (GOV.UK, 14 October 2026). The one dated AI commitment in the press release, an AI Medicines Development Action Plan, is due in 2027 and concerns medicines development, not clinic software.

On deployment, the explainer offers a single phrase: a "whole-system approach to safe deployment", with no detail on what that involves (MHRA Strategy 2030 MedTech explainer, GOV.UK, October 2026). It also says Yellow Card reporting will become easier for healthcare professionals to use, so concerns about devices reach the regulator faster.

One limit on this reading: we read the GOV.UK press release and the MedTech explainer, not the full strategy PDF. The absence of scribes is a statement about those two documents, not a claim about every page of the strategy.

So what for you: read the strategy as a regulator describing how it will supervise manufacturers. It creates no new duty for a practice that uses a scribe, and it should not move your policy this month.

The three dates, and who each one is for

The first is December 2026. The MHRA will issue draft guidance on managing changes to AI-enabled devices by then (GOV.UK, 6 October 2026). This is aimed at manufacturers who update models after they are on the market. It matters to you indirectly, because a scribe vendor that changes its underlying model without telling customers is the situation this guidance is meant to discipline.

The second is spring 2027. The government accepted all 44 recommendations of the National Commission into the Regulation of AI in Healthcare, whose report appeared on 10 September 2026, and promised a full implementation roadmap with timelines and responsibilities by spring 2027 (GOV.UK, 6 October 2026). The Commission was chaired by Professor Alastair Denniston and, according to pharmaphorum (October 2026), drew on input from more than 12,000 patients, clinicians and members of the public. Until the roadmap arrives, "accepted" tells you the direction, not the obligations.

The third is 2027, with no month given. The MHRA will begin consulting on how to qualify and classify AI-enabled devices (GOV.UK, 6 October 2026). This is the date that could matter to a practice, because qualification and classification is where the line between a non-device scribe and a device scribe is drawn. There is also an MHRA AI Airlock Phase 3, with a webinar for applicants on 22 October 2026 and first innovators selected in November 2026 (GOV.UK, 6 October 2026), but that programme is for innovators testing post-deployment monitoring, not for practices.

So what for you: of the three, only the 2027 consultation can change the status of a tool you already use, and it has no published start date. Treat it as a watch item, not a deadline.

The rule that applies today has not moved

The MHRA's guidance on ambient voice technology, issued on 29 July 2026 with NHS England, draws the line by intended purpose. Tools used only for transcription, consultation summaries, draft referral letters or suggested clinical codes for clinician review are not medical devices under current law. Products intended to support diagnosis, treatment or prevention, or that take automated action without clinician review, are regulated as devices, and the clinician remains responsible for checking outputs before they are used in care (Dentistry.co.uk, 2 August 2026, reporting the guidance). Our earlier explainer sets out how that line applies to a psychology practice, including how a vendor's own promotional wording can place a feature on the wrong side of it.

Your obligations therefore sit where they sat on 13 October: a data protection impact assessment under UK GDPR where processing is likely to be high risk (recording therapy sessions with a new technology will usually meet that test), client consent that you can evidence, your regulator's standards and the BPS code, and an indemnity policy that has been told about the tool. None of the October announcements touches any of them.

Here is where the 2027 consultation could bite, and this is our reading rather than anything the MHRA has said. If the qualification test tightens, the features most likely to be reclassed are those that go beyond writing up the session: automated risk flags, "generated insights" or suggested formulations sitting alongside the note. If that happened, your vendor would carry the conformity burden, and you would carry the decision about whether to stay on a product that had not been through it.

So what for you: choose and configure your scribe on its narrowest function today. A tool that transcribes and drafts, with every output signed off by a clinician, is the version least exposed to whatever the 2027 consultation decides.

What 44 accepted recommendations leave unsaid for a private practice

The government response sets out ten priority areas and says it will work with health system partners to improve access to redress when care falls below expectations. Professor Henrietta Hughes, Deputy Chair of the Commission and Patient Safety Commissioner for England, called for "clear routes to redress when things go wrong" (GOV.UK, 6 October 2026). As we read the release, it does not address who is liable when a clinician relies on an AI output, it does not mention independent providers, and it does not mention the CQC.

That silence matters because private psychology practice sits outside most of the NHS-facing machinery the Commission was built around. Until the roadmap assigns responsibilities, the working assumption is the same one that applied last month: the registrant who signs the note owns the error, and their indemnity cover is what responds. Our piece on professional indemnity and AI scribes covers the disclosure questions insurers are now asking.

So what for you: do not wait for the spring 2027 roadmap to settle who is accountable. Check now that your own AI policy says the signing clinician reviews and owns every AI-drafted entry, and that your insurer has seen it.

A 30-minute check for this week

Four tasks fit inside half an hour. First, ask your scribe vendor in writing for its stated intended purpose, and file the reply with your DPIA, because intended purpose is the test the July guidance applies. Second, ask how and when the vendor notifies customers of model changes, since the December draft guidance is aimed at exactly that behaviour. Third, ask whether any feature produces risk flags, insights or suggested diagnoses, and switch off or ring-fence anything that does until you have read the 2027 consultation. Fourth, put two entries in the practice governance calendar: a review in December 2026, when the draft change-management guidance lands, and another in spring 2027, when the roadmap is due.

The call: do not rewrite your AI policy this month. Frankly, a policy that already names the tool's purpose, the DPIA, the consent route and the signing clinician is in better shape than the headlines suggest, and a redraft written before the 2027 consultation would be out of date within a year. Add the two review dates and the vendor letter, and leave the rest alone.

Where a practice would like an independent read of its scribe set-up against these dates, that is what the AI Opportunity and Growth Assessment examines, and a free 20-minute discovery call is the place to start.