NHS AI triage tool goes live in 12 hospitals

The NHS deployed its long-awaited AI triage tool to 12 hospitals this week, after a year of pilots and regulatory back-and-forth. The tool triages A&E patients based on symptoms and vital signs.

The deployment marks the largest non-trial AI deployment in NHS history. The 12 hospitals cover a catchment of 4.5 million patients and see roughly 600,000 A&E attendances per year combined.

The tool, developed by a UK startup called AIVaid, with funding from NHSX and the Wellcome Trust, analyses patient-reported symptoms and vital signs (pulse, blood pressure, oxygen saturation) to assign a triage priority.

How it works:

Patients arrive at A&E and are assessed by a triage nurse as usual. The nurse enters the data into the NHS system. The AI tool runs in the background and provides a recommended triage priority (immediate, very urgent, urgent, standard, non-urgent) within 30 seconds.

The nurse sees both the AI recommendation and a confidence score. The nurse makes the final triage decision. The AI recommendation is logged but not auto-applied.

Why this matters:

A&E triage is a high-stakes decision. Under-triage risks patient harm. Over-triage wastes resources. The AI tool claims to reduce under-triage by 23% while maintaining or improving over-triage rates in pilot data.

The deployment is heavily regulated. The MHRA classifies the tool as Class IIa medical device, requiring post-market surveillance and adverse event reporting. The NHS has set up a dedicated oversight committee.

Criticisms:

The tool has critics. The British Medical Association has raised concerns about:

• Liability when the AI is wrong.
• The training data being skewed toward English-speaking patients.
• The potential for automation bias (nurses over-trusting AI recommendations).
• The lack of patient consent for AI involvement in their care.

The NHS has responded with mitigations:

• Clear liability allocation (nurse is responsible for triage decision).
• Multilingual support with translation.
• Mandatory training on automation bias.
• Patient information leaflets explaining AI involvement.

The deployment will be evaluated over 18 months. If the outcomes hold up — if the tool genuinely improves triage without introducing new harms — it could become the model for AI deployment across the NHS. The UK would then be a global leader in clinical AI, not just consumer AI.