AI-Enabled Ambient Scribing in Healthcare: New Guidance

Executive Summary

New NHS England guidance on AI-enabled ambient scribing products used in health and care settings sets out a detailed framework on how to implement such products safely and responsibly. This is essential reading for both private and public sector organisations who are considering, or have already deployed, such technologies.

The document aligns with MHRA guidance that AI transcribers will be medical devices where they produce summaries which are used for a medical purpose, meaning such devices will need the appropriate certification. 

Background 

On 27 April, NHS England released comprehensive guidance on the use of AI-enabled ambient scribing products in health and care settings. These tools use advanced speech technologies including Large Language Models to automatically convert conversation between patients and clinicians into clinical documentation, and are intended to streamline documentation and workflows accordingly. This guidance is designed to support NHS organisations in the safe implementation of these tools. Given the sensitive nature of healthcare data and the embedded data protection implications, robust safeguards are essential to ensure the confidentiality and security of patient information.

We summarise key implementation steps, regulatory compliance points and potential benefits of such tools for NHS organisations to consider below.

Functions and purpose of AI-enabled ambient scribing products

AI-enabled ambient scribing products are designed to improve patient and clinical documentation and workflows. NHS England claim that these tools “unobtrusively record conversations between patients and caregivers” and automatically convert spoken words into structured medical documentation with minimal user intervention.

Key functions include: 

• capturing and recording speech interactions; 
• incorporating additional context from health records; 
• generating outputs such as summaries or medical letters; 
• formatting outputs according to predefined templates; 
• extracting and linking terms to clinical codes;
• populating health records; and 
• suggesting follow-up actions such as scheduling or referrals. 

The primary purpose is to “enhance clinical efficiency, improve patient care, and reduce clinician workload”. A key intention is in ensuring that caregivers and clinicians are able to focus more on patient interaction rather than administration.

We have seen some examples where practitioners seek to avoid such tools gathering personal data, such that the regulatory requirements applicable to healthcare data do not apply. However, as these products evolve it seems increasingly likely that they will need to hold personal data in order to be used most effectively. 

Key implementation steps and regulatory considerations

The guidance outlines several steps and regulatory considerations for NHS organisations looking to implement AI-enabled ambient scribing tools. 

The guidance aligns with previous MHRA guidance on AI transcriber and consultation summarisers which stated such software would be considered a medical device where it summarises consultations which are then inputted into a patient’s electronic health record (see MHRA guidance Digital Mental Health Technology - Regulation and Evaluation for Safe & Effective Products for further information). 

The result of this is that any AI-enabled ambient scribing products which produce summaries, or any output other than a verbatim transcript, which are used by healthcare professionals for any medical purpose (e.g. if they become part of a patient record and/or are consulted to determine any diagnosis, monitoring, treatment or prevention) will be medical devices which require the appropriate certification.

Organisations must: 

• Appoint a ‘Clinical Safety Officer’ to oversee the implementation process and identify technical risks and clinical hazards;
• Complete certain risk management documentation including developing and maintaining a safety case and updating a hazard log per DCB0160 (the standard for health IT software) and conducting a data protection impact assessment  to determine if data security measures (such as end-to-end encryption and access controls) should be implemented; and 
• Inform patients about tool usage and their data to meet ethical standards on consent, transparency, and accountability.

Additionally, amongst other recommendations, organisations should: 

• Implement and maintain robust data protection, confidentiality and security controls to comply with legal and regulatory requirements for the procurement and deployment of the tools, such as compliance with the Data Security and Protection Toolkit, the UK GDPR and relevant healthcare regulations; 
• Develop and implement a monitoring and maintenance framework that consistently evaluates the performance of the tool (particularly for bias mitigation), audits clinical documentation, and reviews incident reports. This framework should also facilitate updates based on user feedback or to comply with legal and regulatory requirements;
• Prepare for integration with existing IT systems such as electronic patient records, to maintain workflow continuity and ensure users are appropriately trained; 
• Establish clear and comprehensive contracting arrangements with suppliers, as NHS organisations may still be held liable for any claims arising out of the use of AI products, particularly if it concerns a non-delegable duty of care between the caregiver and the patient; and
• Register the tool with the Medicines and Healthcare Products Regulatory Agency and obtain the appropriate regulatory certification if it qualifies as a 'high functionality medical device' due to the use of generative AI for further processing (such as summarisation).

Benefits of ambient scribing products

AI-enabled ambient scribing products are designed to improve clinical decision making and enhance patient care. Notable advantages include: 

• Reduced administrative workload to improve patient engagement and cost savings;
• Increased operational efficiency by minimising documentation and shortening appointment times; 
• Potential for scalability and interoperability with existing patient record systems and workflows; and
• Capture and transcription of real-time data to improve accuracy of patient records. 

Takeaways

By adhering to the recommended implementation steps and guidance set out by NHS England, NHS organisations are expected to improve clinical documentation, enhance patient care, and streamline operational workflows.

Further documents from NHS England will be published over the next six months to aid adoption, including template hazard logs, safety cases, data protection impact assessments, and evaluation guidance.

This development aligns with a broader policy trajectory in which the UK Government is increasingly recognising the transformative potential of artificial intelligence and data-driven technologies within the healthcare sector. 

Notably, this shift was underscored by the publication of the Sudlow Review in November 2024, which set out a compelling vision for a more integrated and accessible health data infrastructure and of which we previously commented on here. In response to these recommendations, the Government has since launched the Health Data Research Service, a centralised platform designed to streamline access to national health datasets. 

This initiative is intended to accelerate medical research, support innovation, and ultimately improve patient outcomes by enabling researchers and clinicians to harness the full potential of the UK’s rich health data resources. 

We are closely monitoring these developments, particularly in relation to data protection, governance, and the evolving regulatory landscape surrounding the secondary use of health data.

If you have any questions or would otherwise like to discuss how the recommendations may impact you or any other issue raised in this article, please contact Rory Trust, Patrick Parkin, Madelin Sinclair McAusland, Amanda Leiu  or a member of Burges Salmon's data team. 

This article was written by Yadhavi Analinkumar and Victoria McCarron.