180 incidents on record · 2026 Headlights Incident reports by Ellie Harris · Melbourne
10 new this week Library last updated 30 August 2026
← The incident library
HD-INC-145
Healthcare · United Kingdom · 2020 · Unsafe triage advice from a health chatbot

An NHS-partnered health app's AI checker was flagged for missing serious conditions, and the regulator shared the concerns

By Ellie Harris · Filed Concerns raised from 2017; MHRA letter reported December 2020

Alleged: Babylon Health developed or deployed the AI system implicated in this incident. Details are drawn from public reports; parties are presumed innocent of any wrongdoing not established by an official finding.

An NHS-partnered health app's AI checker was flagged for missing serious conditions, and the regulator shared the concerns

What happened

It was reported that Babylon Health, whose AI symptom checker was used alongside parts of the NHS in England, faced criticism over several years after concerns were raised that it could miss serious medical conditions.

Consultant physician Dr David Watkins publicly tested the chatbot from 2017 onwards and argued that, in some cases, it failed to recognise symptoms that could indicate life-threatening conditions, including heart attacks and blood clots. Several of his concerns related to how the system assessed chest pain. Babylon disputed his findings at the time and criticised aspects of his testing.

The MHRA, the UK’s Medicines and Healthcare products Regulatory Agency, wrote to Dr Watkins in December 2020, in a letter later reported by TechCrunch, saying his concerns were valid and that the regulator shared them. The letter also highlighted a gap in how AI triage chatbots were regulated, while noting that confidentiality prevented the MHRA from commenting on any specific investigation.

Babylon said its symptom checker met the regulatory requirements that applied at the time, was safe to use, and represented only one part of its wider healthcare business.

In 2023, Babylon entered insolvency proceedings. Its UK clinical services were later sold and continued under new ownership.

What an auditable version would have shown

An auditable triage system would show how well it performs on the conditions it simply cannot afford to miss. It would record how often the chatbot correctly identified symptoms of serious conditions such as heart attacks, strokes and blood clots, how often it failed, and whether that performance changed over time. It would also keep a record of each significant safety concern raised after launch, what investigation followed, what testing was carried out, and whether any changes were made before the system continued to be used. That gives hospitals and regulators evidence of how the system performs in practice, rather than relying on marketing claims.

Where the gap was

The problem wasn’t simply that concerns were raised. The problem was that credible safety concerns about missed serious conditions remained part of a public dispute for years. A symptom checker exists to identify the patients who need urgent medical attention. If evidence suggests it may be missing those patients, that evidence should trigger independent testing and review as quickly as possible. At the time, the regulatory framework for AI triage systems was still developing, leaving uncertainty about how those concerns should be independently assessed.

What governance should have looked like

A symptom checker should be judged first on the conditions it cannot afford to miss. Before deployment, and throughout its life, it should be regularly tested against high-risk clinical scenarios such as chest pain, stroke symptoms and suspected blood clots. If credible evidence suggests its performance has fallen below an acceptable standard, that should automatically trigger further testing and, where necessary, changes before wider use continues. A VerificationGate is designed to ensure a safety-critical system passes those high-risk clinical tests before deployment and after significant updates. A MetricRecord keeps a transparent record of how often the system correctly identifies, misses or incorrectly classifies serious conditions. The goal isn’t to prove an AI system is perfect. It’s to make sure that decisions about patient safety are based on evidence that regulators, clinicians and the public can independently examine.

The reference implementation of VerificationGate and MetricRecord is open source. It lives at github.com/saffronandindia/headlights-oss, Apache 2.0 licensed, free for any company to install. The repository is public now.

Sources

The mailing list

Fresh incident reports every week. One email to match.

We add new incidents to the library regularly, and send a single short email each week with what's new. The library stays free and open; this is just how you keep up with it.

No tracking. Unsubscribe in one click.

The record

An auditable system would have produced a signed, tamper-evident record the moment this happened: what the system did, the version that did it, the basis it acted on, and the action taken, and Babylon Health could have produced it on demand.

This is the record the system as deployed did not produce in a signed, auditable form.

What this teaches
Capture what happened when it happens
What the system did, the version that did it, the basis it acted on, and the action taken, recorded at the moment, not reconstructed after.
Sign it, so no one has to trust the record-keeper
A tamper-evident entry. Edit it later and the signature breaks. The record does not ask for the benefit of the doubt.
Make it verifiable by anyone
A court, a regulator, a customer's lawyer can check the record themselves, without taking the company, or us, at our word.

Headlights summarises publicly reported AI incidents. All summaries are independently written, attributed to their original sources, and intended for research and educational purposes. Allegations are identified as such until established through official findings.

This report is based on the sources listed above and reflects information available at the time of review; later developments may not be captured. Where a person is described as charged with or alleged to have done something, that allegation is unproven unless a conviction or a court or regulatory finding is stated. Headlights publishes journalism and commentary, not legal advice.

Want to write back?

Direct to my inbox.

ellie@useheadlights.com →