The National Commission into the Regulation of AI in Healthcare report, published on 10 September 2026, told the government what The Pharmaceutical Journal told pharmacists more than 18 months ago: healthcare professionals still need clarity on their responsibilities when using artificial intelligence (AI).
In that time, multiple pieces of guidance on AI for pharmacists have been published. But none have established protections for errors and biases that originate from behind the AI tool, rather than from the actions of the pharmacist using it.
AI’s promise, as The Pharmaceutical Journal put forward in February 2025, is to ease pressure on an overstretched workforce — particularly in relation to the administrative and research tasks covered in our recent learning article on AI. However, as a naturally risk-adverse profession, pharmacists have reason to be wary when that promise meets patient-facing decisions. A pharmacist pushed to adopt these tools to save time should not also pay the price when a flawed or biased training dataset — built by someone else — goes wrong.
Pharmacists are potentially putting their professional standing on the line for a rapidly evolving technology
It’s true that pharmacists should be accountable for the decisions they make in the pharmacy. This is clearly iterated in each of three major pieces of guidance from the then Royal Pharmaceutical Society (now the Royal College of Pharmacy [RCPharm]), the General Pharmaceutical Council (GPhC) and the Pharmacists’ Defence Association. They say the same thing in different words: pharmacists “remain accountable for professional decisions in the pharmacy”; they “remain personally accountable for their decisions and actions”; accountability for AI-influenced decisions “stays with the pharmacist”.
But what happens if bias and errors are built into tools before they ever reach the dispensary? For example, a training dataset that under-represents a population, a hallucinated drug interaction, or an answer stitched together from open-web content, a paywalled abstract or other unverified data that is not held to the evidence bar required by clinical decision-making.
Currently, as the NHS Race and Health Observatory’s (RHO) briefing on AI — published the same day as the National Commission’s report — puts it: “The risk sits with the clinician using the tool, not the manufacturer” that built it.
This means pharmacists are potentially putting their professional standing on the line for a rapidly evolving technology that has the scope to be so dangerous that experts have said a ‘kill switch’ should be mandated.
The results of RCPharm’s own survey of 141 pharmacists revealed this to be a significant concern: 74% (n=104) said AI in practice gave them some cause for concern, and among those with concerns, 92% (n=96) pointed specifically to the risk of error or hallucination — exactly the type of harm that originates in the tool itself.
Tase Oputu, president of RCPharm, has commented that the risk posed by using AI should be allocated according to who is best placed to control it, “rather than defaulting to frontline clinicians” as it currently does.
In the same way that a pathway exists for reporting, escalating and withdrawing medicines that cause harm — the Yellow Card scheme — a similar pathway should exist for AI systems used in healthcare
The National Commission’s report seeks to remedy this by calling for a coordinated approach, across the Department of Health and Social Care, devolved health departments, regulators and royal colleges, to developing the use of AI in healthcare — covering education, training and professional development — and, crucially, for contracts between AI device manufacturers and healthcare providers to include an explicit allocation of responsibility for controlling risk.
We welcome the publication of a joint statement of intent — from the Professional Standards Authority for Health and Social Care and professional regulators, including the GPhC and the Accredited Registers Collaborative — committing to developing principles to guide how they will regulate the use of AI by healthcare professionals across the UK, but additional action and rapid progress is needed now.
As a royal college, RCPharm can and should press for that allocation of responsibility to be written into pharmacy’s own contracts, develop model contract clauses for employers procuring AI-enabled dispensing and decision-support tools, and work with the GPhC and others to spell out what “accountable” looks like once a third-party system is involved.
In the same way that a pathway exists for reporting, escalating and withdrawing medicines that cause harm — the Yellow Card scheme — a similar pathway should exist for AI systems used in healthcare, to help guide pharmacists on what systems should be avoided in practice. This was the RHO’s recommendation as it points to pulse oximeters calibrated on light skin tones overestimating oxygen readings in patients with darker skin, as a warning of the kind of harm that travels furthest when no one is watching for it.
No one has yet drawn a clear line between what a pharmacist should answer for and what a manufacturer should answer for, and the evidence to draw one is sparse. No pharmacist in the UK has yet faced a reported clinical negligence claim over AI use. The nearest precedent is borrowed from law, where the High Court criticised lawyers in 2025 for submitting AI-hallucinated case citations without checking them first.
RCPharm, the GPhC and the PDA should now push to draw that line of accountability between pharmacists’ clinical judgement and AI developers, or the risk will remain
It is easy to conclude that the fix is simple: pharmacists should just check what the AI tells them. The pulse oximeter example shows why that falls flat. Each miscalibrated reading looks plausible and clinically sound on its own, with the bias only becoming visible in the aggregate. An AI hallucination works the same way — the dangerous ones are the ones that look right.
RCPharm, the GPhC and the PDA should now push to draw that line of accountability between pharmacists’ clinical judgement and AI developers, or the risk will remain, with a pharmacist facing the consequences of a mistake that they personally did not make. PJ
