The Royal College of Physicians (RCP) has expressed concern following the UK government’s announcement that the Medicines and Healthcare products Regulatory Agency (MHRA) will not classify all automated voice technologies (AVTs) used in healthcare as medical devices.
New guidance from the MHRA has clarified that AVTs intended ‘solely for transcription, summarising clinical conversations, drafting letters or suggesting clinical codes for clinician review‘ are not regulated as medical devices under the current framework, and that this will not change. However, this marks a departure from previous NHS guidance, under which suppliers needed to hold MHRA Class I medical device registration to be included on the NHS-approved AVT self-certified supplier registry.
AVTs are increasingly being used across the NHS to record, transcribe and summarise clinical consultations. However, these tools do not simply provide voice transcription. They use generative AI to identify, interpret and prioritise information, determining what is important enough to include in a summary. In this sense, they function as ‘inference machines’, applying automated judgements about relevance, significance and meaning that can influence clinical decision making.
The RCP has consistently called for smarter regulation that supports innovation and rapid adoption while increasing clinician and patient confidence in these technologies. By not including all AVTs used in healthcare from the medical device regulatory framework, this could weaken oversight of technologies that provide clinical decision support, making it harder to identify and respond to risks. In addition, where an AVT produces inaccurate documentation, omits relevant information or contributes to patient harm, responsibility will fall more heavily on individual clinicians and trusts, rather than being shared within a regulated device framework, as reported this week in the HSJ.
The new guidance places significant weight on the stated intended purpose of manufacturers when determining whether a product is regulated as a medical device. This could create a scenario where similar technologies may be regulated differently depending on how they are marketed, rather than how they function in clinical practice.
The MHRA has said that classification depends on the intended purpose of the software rather than the fact that it is voice-based technology, and that tools which simply record or transcribe consultations fall outside device regulation. This distinction is not straightforward in practice. Where AVT-generated summaries inform diagnosis, treatment decisions or coding, it is difficult to argue that the software is not providing decision support in all but name.
Dr Anne Kinderlerer, RCP clinical digital health lead, said:
‘We are concerned that not including all automated voice technologies in healthcare settings from the medical device framework risks creating gaps in oversight at a time when these tools have the potential to influence decisions about patient care. If an AI-generated summary omits clinically relevant information or introduces errors, the consequences can extend far beyond simple notetaking.
‘Under-regulating AVTs now risks a much larger correction further down the line. Any failures or patient safety incidents could lead to a loss of confidence among healthcare professionals and the public, followed by the need for more restrictive intervention later. Getting the regulatory approach right from the outset is essential.
‘The RCP’s position has always been that the maximum benefit of AVTs lies in their integration with electronic patient records and their ability to operate as part of a genuinely agentic AI system. These tools could support clinicians to order the right tests, document information appropriately, and generate structured, shareable coded data. By stepping back from regulating all AVTs in healthcare as medical devices, the MHRA risks locking us into a fragmented landscape of tools that cannot integrate effectively. That is not the future of safe, effective digital care that patients and physicians need.’
The RCP is calling for mechanisms for continued oversight of these tools with robust benefits evaluation and systems to detect harm. The college will continue to engage with the government, regulators, NHS England and industry to advocate for a framework that supports safe, integrated and interoperable AI in clinical practice.