The NHS is betting on AI to cut paperwork. Its own watchdog says patients are paying the price

She was told her MRI showed nerve damage. The AI had heard wrong.

The hospital had not given her a diagnosis. An AI scribe (software that listens to a consultation and drafts the clinical note) had misheard "null demyelination," a clean scan, as "demyelination," the term for serious nerve damage linked to multiple sclerosis. The note went into her record. The patient, herself an NHS health professional, was the one who flagged the error. The hospital eventually corrected it.

"It was a very traumatising experience to be given an incorrect diagnosis because of AI and then be told it's a typo," she told Healthwatch England, the statutory patient champion for the NHS in England, which published the case as part of a new warning about AI scribe accuracy.

That warning, reported by The Guardian, describes a pattern. Healthwatch says it has heard "multiple stories from patients who have noticed these errors when a health professional hasn't." In one case, an AI scribe confused a GP-prescribed drug with a different one of similar name; the patient, not the doctor, caught the mistake. In another, an AI-generated summary letter failed to record that a hospital consultant had told the patient to obtain a repeat prescription from their GP for migraine medication, a dropped instruction that risked an interruption of treatment.

The errors share a single shape. The AI mis-transcribes a consultation. The note is filed. The clinician signs off. The patient reads the letter weeks later and notices something wrong. The cost of correction falls on the patient.

"Healthcare has never been error-free," a Healthwatch spokesperson said. "But our findings show the urgent need for clarity over how patients can report and get corrected any mistakes made by AI scribing tools or the professionals that use them."

That ask is the constructive core of the warning. Healthwatch is not calling for AI scribes to be rolled back. It is calling for a clear correction route, one that, on the current evidence, does not exist as a patient-facing service. Rachel Power, chief executive of the Patients Association, put the same point differently: "Trust and confidence in this technology depend on good communication and genuine partnership with patients, and right now both are missing."

The gap matters because AI scribes are no longer a pilot. They are central to the UK government's 10-year health plan for the NHS in England, which positions the tools as the mechanism for a "big shift" from analogue to digital services and expects them to "liberate staff from their current burden of bureaucracy and administration." NHS England is scaling deployment across trusts. One NHSE region has signed an AI scribe deal covering 15 trusts, and a South West London rollout of vendor Lyrebird Health is described as the UK's largest ambient AI implementation, reaching roughly 20,000 clinicians. The Concordance tracker lists the wider deployment footprint.

Regulators are catching up. The MHRA and NHS England have issued clarification on the rules for AI clinical scribes, indicating an active but unsettled perimeter around how the tools are governed, what they are allowed to draft, and who carries professional responsibility for what is filed.

The structural question, then, is who is the last line of defence when an AI scribe gets the record wrong. The Guardian's report makes the answer explicit: the patient, reading their own letter after the fact, with no clear escalation route and no contractual duty on the clinician to verify the AI's output before signing.

Healthwatch's ask is the smallest unit of trust infrastructure the rollout needs: a clear reporting and correction channel, built before the next wave of deployment. Without it, every patient is being asked to act as the verification layer for a system that, by design, files its work before anyone reads it.

The next test is whether the NHS builds that channel, or whether the next patient who reads their own letter will be the one to notice the mistake.