Artificial intelligence in primary care: navigating the medicolegal implications

This is part of the Pulse Partners series. This article has been paid for by the Medical Defense Society, with editorial input by Pulse. The opinions in this article do not necessarily reflect the views of Pulse.

For many GP practices, stretched by rising patient demand and administrative burdens, artificial intelligence (AI) tools offer a glimmer of hope. Drawn by the promise of more efficient workflows and improved access and care for patients, some practices are enthusiastically adopting new AI technologies.

Yet others remain cautious. Many GPs we speak to are wary about the medicolegal, ethical, and regulatory challenges AI may bring, while practices in more deprived areas may lack the resources to explore its potential, raising questions about inequities in adoption.

Despite these concerns, the introduction of AI into everyday practice is actively encouraged by government policy, including the 10-year health plan. As AI technologies develop further, they are likely to become embedded in many aspects of general practice.

Given their rapid evolution, it is unsurprising that so far, implementation in primary care remains patchy. Nevertheless, over a quarter of GPs responding to the Royal College of General Practitioners GP Voice Survey 2025 reported using AI in their practice.

In that survey, GPs’ priorities for AI centred on automating administrative and documentation tasks and supporting patient education. These functions could reduce workloads and free clinicians to focus on more complex aspects of care, with research suggesting significant potential for time and cost savings.

The challenges and medico-legal implications

A growing range of tools is being developed to support both clinical and administrative work in primary care and at the interface with secondary care. However, use varies widely between practices, and the potential benefits may not always be realised.

One challenge is that clinicians are not always clear what counts as AI, which systems incorporate it, or how to interpret outputs. The complexity of AI systems often makes reasoning opaque and results difficult to verify. For example, AI results may be biased if trained on skewed data, yet individual clinicians usually cannot check this. This ‘black box’ nature introduces medico-legal risks, which may hinder wider adoption in healthcare settings.

Our members frequently ask how to implement AI safely, balancing the benefits against new medico-legal liabilities. Picture the following scenarios, in which a practice and its GPs interact with AI tools. These illustrate some of the risks that could arise and the safeguards to protect against them.

A practice’s AI-enabled triage system directs a patient to a routine GP appointment in a week’s time. However, the patient actually needs urgent care and later contacts the out-of-hours service in distress.

The practice introduced the AI triage system to improve access and meet NHS England requirements to make online consultation tools available during core hours. However, AI triage decisions depend on the data patients provide, and misclassification of symptoms by the AI in this case creates a patient safety concern and medico-legal risks for the practice.

This demonstrates why AI triage tools need to be carefully designed for safe navigation by patients. Safeguards must be built in to prevent misprioritisation of requests by AI systems in the absence of clinical oversight, and practices must ensure that clear messaging directs patients to contact them by phone or in person if their condition is urgent.

Dr F uses an AI clinical decision support system to guide treatment. If a patient is harmed, who is liable: the AI developer, the NHS, the practice, or Dr F?

These tools aid clinical decisions by identifying medical conditions, suggesting treatments or flagging risks, but they are fallible. The General Medical Council (GMC) guidance on the use of AI and innovative technologies is clear: doctors remain responsible for decisions influenced by AI. Dr F should follow Good Medical Practice, discuss the use of AI with his patients to gain their informed consent, verify the medical information that is generated, and act on any safety concerns. He should be aware that if he makes a clinical decision based on inaccurate advice from the AI tool, he may be held liable for any claims.

Most AI clinical decision support tools are classified as medical devices and regulated accordingly. Using them outside their approved scope may increase personal liability. We advise doctors to apply AI according to its intended and approved use and within their area of clinical expertise. They should be ready to override inaccurate results, and must document decisions and consultations, being prepared to justify their clinical reasoning.

Dr M uses AI transcription software to generate consultation notes. What medico-legal risks does this involve and how can they be managed?

AI scribes using ambient voice technology (AVT) are increasingly promoted to improve efficiency by transcribing consultations into documentation, including patient records. But while they may free clinicians to focus more on patient interaction and potentially improve the quality of records, they often use generative AI and must be used with care.

In this scenario, Dr M remains responsible for the accuracy of the medical record and could be liable for errors generated by the software. She should therefore review and correct AI-generated notes before finalising them, with particular care in complex consultations or where language barriers exist.

Practices are responsible for the safe implementation of these products and must follow NHS England guidance on the safe adoption of ambient scribing products. This means ensuring appropriate training, compliance with medical device regulations, secure integration with existing digital systems, and adherence to information governance requirements.

Transparency with patients is also important: practices and clinicians should inform patients that AI tools are being used and allow them to object. The Care Quality Commission advises that explicit consent for use of AI scribes is not usually needed, as clinicians may rely on implied consent under the common law duty of confidentiality. However, the type of consent required (implied or explicit) depends on the type of AI technology and how it is used.

Dr G’s patient needs a referral to secondary care, and she uses AI to draft the referral letter. What are the medico-legal implications?

AI scribing tools using AVT and generative AI may also be used to draft referral letters for secondary care, relieving GPs of a significant administrative burden and potentially enhancing the speed and accuracy of referrals. However, Dr G remains responsible for the content and must review and correct any errors before approving the letter.

She should be aware of the risk of compounding errors, where incorrect or incomplete AI-generated information is used by the receiving clinicians to influence their decisions. The errors consequently become baked into the patient’s care plan, increasing medico-legal exposure for all concerned.

If the tool embeds triage urgency flags or SNOMED-coded data in referrals, Dr G must verify their accuracy to ensure appropriate prioritisation and continuity of care. As the referring clinician, she would be liable for any claims resulting from inaccuracies.

In addition, because referral letters are transmitted between digital systems of different care settings, AI tools for this purpose must be carefully implemented to ensure compatibility, secure data transfer, and protection of patient privacy, following the NHS guidelines.

Dr T is interested in using a publicly-available generative AI tool to help draft correspondence with patients and secondary care. What are the potential pitfalls?

Generative AI systems based on large language models can produce text from prompts, which could help Dr T to draft responses to patient queries or correspondence with secondary care specialists. However, using an unvalidated tool outside approved NHS systems would raise concerns about information governance and data security and is not recommended.

A major legal risk for Dr T is a data breach if patient information is entered into a public AI platform. Even without names or addresses, details about rare conditions or unique combinations of diagnoses may still make it possible to identify patients.

The information generated by the AI tool could also create a potential pitfall, as such systems may produce incorrect references or ‘hallucinations’ and reflect biases in their training data.

Managing risk when introducing AI

In our experience, GPs are keen to adopt technology that reduces administrative burden but they often seek reassurance about safe implementation and minimising medico-legal exposure. Particular concerns relate to issues including clinical accountability, reliability of AI outputs, record-keeping responsibilities, consent and transparency, information governance, regulatory compliance and the impact on workload and safety.

We advise that before implementing AI tools, GPs and practice leaders gain an understanding of the regulatory requirements, establish clear governance processes to manage risk, and recognise that clinicians remain responsible for clinical decisions and therefore liable in the event of claims.

Key steps for managing the risks include:

• Procuring compliant systems: Ensure AI tools meet regulatory standards such as DCB0160, Digital Technology Assessment Criteria (DTAC), and Medicines and Healthcare products Regulatory Agency registration where classified as medical devices. Use of non-compliant tools may place liability on the organisation or individual clinician.
• Ensuring oversight: Appoint a clinical safety officer and digital lead with appropriate Digital Clinical Safety training to oversee risk management.
• Assessing risks: Conduct clinical safety risk assessments and data protection impact assessments, and maintain a hazard log.
• Clarifying contracts: Protect practice interests by ensuring that contract arrangements with AI suppliers are clear and comprehensive in detailing roles, responsibilities and liabilities.
• Protecting patient data: Implement information governance policies ensuring compliance with regulations such as UK General Data Protection Regulation.
• Training staff: Provide appropriate training for all users.
• Obtaining appropriate indemnity: Ensure that indemnity arrangements for practices and GPs cover their use of specific AI-enabled tools.

As noted in the NHS guidance on ambient scribing products, liability for claims arising from the use of AI products in the NHS remains largely untested, as few legal cases have been brought to date. Where responsibility cannot be clearly attributed to a specific party, liability may rest with the NHS trust or primary care provider, which retains a non-delegable duty to ensure patient safety and the quality of care.

It is important to understand that in certain scenarios, particularly where professional judgement or regulatory expectations are not met, the use of AI may not be fully covered under standard indemnity arrangements.

Therefore, when considering introducing a new AI system in practice, we recommend that GPs and practice leaders contact their indemnity providers for a detailed discussion about the potential liabilities and mitigations, and to ensure they are fully protected. It is always best to seek clarity and examine use case scenarios before accidentally straying into a grey area.

AI tools have the potential to support general practice in many areas, including triage, clinical decision support, and transcription. However, the scenarios and risks we describe above underline the importance of seeking early advice from indemnity providers, ensuring robust governance arrangements, and maintaining clinical oversight when adopting AI technologies. Cautious implementation with adherence to NHS and GMC guidance may allow GPs to harness the benefits of AI while minimising risks to their patients and themselves.