An Absence of Reported Incidents Is Not Evidence of Safety

An absence of reported incidents is not evidence of safety. It can simply be evidence that nobody knows what to report, or where to report it.

Yesterday the Health Services Safety Investigations Body (HSSIB) launched a patient safety investigation into ambient voice technology in hospitals. Alongside patient harm and safety risk, one observation in their intelligence review stood out to me: the routes for recognising and reporting AI-related incidents may not yet be robust enough for emerging risks to surface at all.

That is a measurement problem sitting underneath a safety problem.

A Concept Prototype

Two months ago I published an open concept prototype testing one answer to it: a UK Healthcare AI Safety Registry. The design assumptions are in the open for anyone to pull apart.

• A publicly searchable record of risks, near misses and harms from healthcare AI deployments
• Every entry classified against the MIT AI Risk Repository taxonomy, so hazard analysis under DCB0129/0160 draws on one shared vocabulary instead of two hundred local ones
• A two-field report a clinician can file in ten seconds at the end of a busy clinic - product name and severity, everything else optional

To be clear about what it is not: it is a concept prototype populated with illustrative data. Not a live service, not endorsed by anyone, and not something HSSIB has asked for.

What I'd Value Most Right Now

Challenge. If you work in patient safety, clinical safety or informatics: what is missing? What would stop you using it? And is a national registry even the right shape, or does this belong inside LFPSE?

TrustPoint Digital Health provides independent clinical safety governance for AI and digital health products under DCB0129 and DCB0160. Get in touch to discuss your safety case.