ELSA AI · Clinical AI Governance

Where AI exposure hides in a private clinic

Ambient scribes, ChatGPT, Microsoft Copilot, transcription and admin tools are already in use across most clinics, often before any governance position is documented. This map shows the exposure themes a board should understand, and the controls that reduce each one.

This map is illustrative; your clinic’s actual position comes from the Diagnostic.

Threat modelling: STRIDEOWASP LLM Top 10DCB0129 / DCB0160UK GDPR & DPA 2018ICO · CQC · MHRA expectations

Three forces create AI exposure

External threat actors

Targeting patient data, accounts and the vendor supply chain behind AI tools.

Well-meaning insiders

Clinicians and staff using AI to save time, often working around gaps that governance has not yet closed.

The technology itself

AI’s own failure modes: confident but incorrect output (hallucination) and uneven performance across populations (bias).

The map

Exposure themes

Three themes by what is at stake. Open a theme to see how the exposure typically arises and what it means. Priority levels show the general pattern across clinics, not a rating of any single practice.

View
Typical priority
Shadow AI: data entered into consumer toolsHigh priority
How it arises
Staff use unapproved consumer AI tools to draft letters or summarise histories.
What happens
Identifiable health data is entered into public tools and can be retained to train future model versions.
What it means
A confidentiality and data-protection exposure that sits outside the clinic’s assurance position and warrants DPO and legal review.
Cloud processing and data residencyHigh priority
How it arises
Ambient scribe data is processed outside the UK without a documented data processing agreement or transfer mechanism.
What happens
Data is exposed where encryption and access controls are not evidenced.
What it means
A vendor jurisdiction and international transfer point requiring DPO and legal review and vendor confirmation.
Audio retention and over-collectionModerate
How it arises
Ambient voice tools capture consultation dialogue without applying data minimisation.
What happens
Raw audio and verbatim transcripts are kept without documented justification or deletion.
What it means
A retention and minimisation review point. A data protection impact assessment is likely indicated.
Controls

How clinics reduce this exposure

The guardrails that close most of the map. Each shows the practical trade-off across four axes. For risk reduction and readiness, more bars is stronger; for cost and complexity, fewer bars is lighter.

Governance

Impact assessment and clinical safety case

Run a live data protection impact assessment and a local clinical safety case before deployment, and name a trained Clinical Safety Officer.

Risk reduction

High

Cost

Medium

Complexity

High

Readiness

Needs lead-in

Identity & dataQUICK WIN

Close down shadow AI

Block consumer AI tools on clinic devices and offer a sanctioned, private alternative so staff have a safe route.

Risk reduction

High

Cost

Low

Complexity

Low

Readiness

Immediate

Clinical practice

Review before it reaches the record

Require every AI output to be actively reviewed and corrected by the clinician before it enters the patient record, with an auditable marker for AI-assisted entries.

Risk reduction

High

Cost

Low

Complexity

Low

Readiness

Phased

Privacy by design

Delete what is no longer needed

Set tools to delete raw audio and verbatim transcripts once the verified summary exists.

Risk reduction

High

Cost

Medium

Complexity

Medium

Readiness

Phased

ProcurementQUICK WIN

Match the tool to its approved use

Check each tool against its registered intended purpose and rule out using an administrative tool for clinical decisions.

Risk reduction

High

Cost

Low

Complexity

Low

Readiness

Immediate

See where your clinic stands

This map shows common themes. The Clinical AI Exposure Diagnostic™ maps your declared and shadow AI use, your evidence gaps and a 30-day priority action plan, as a fixed-fee review delivered in four working days.

What ELSA AI does not do

  • Provide legal advice or determine legal compliance
  • Give CQC or ICO approval
  • Determine insurer coverage or MDO indemnity
  • Sign off clinical safety cases or author DCB0160
  • Guarantee compliance or a CQC outcome

This resource is an illustrative governance map of common AI exposure themes in private healthcare. It is advisory and does not determine breach, regulatory failure, indemnity position or clinical-safety status for any specific clinic. Final legal, data protection, clinical safety, regulatory, insurer and indemnity decisions remain with the clinic’s accountable officers, including the DPO, legal counsel and Clinical Safety Officer.