Simulator 04 · Live

Hazard Log Builder.

A nine-step rehearsal of the single most important artefact a Clinical Safety Officer produces. Realistic AI deployment scenario, structured feedback at every step, audit-ready PDF export.

Today's scenario

AI Cancer Referral Prioritisation Tool.

An AI tool reads referral text, blood results and prior notes, and labels suspected upper GI cancer referrals as urgent, soon, or routine. The simulator walks you through one defensible hazard log entry for the failure mode that follows when those data sources are not aggregated.

  • 9 structured steps · ~10 minutes
  • Reference answer with concept-level feedback
  • Branded PDF export, audit-ready
Progress0%

Step 1 of 10

Scenario briefing

Deployment

A healthcare organisation deploys an AI tool to prioritise suspected upper GI cancer referrals. The tool ingests the referral text, recent blood results and previous clinical notes, and labels each referral as urgent, soon, or routine.

Safety event

A patient with weight loss, anaemia and intermittent dysphagia is labelled routine. The relevant red-flag features are present in the record but split across the referral letter, the blood results and the prior clinic notes. The model fails to aggregate them, so the case never reaches the urgent queue.

What you'll work through

  • Distinguish a clinical hazard (patient-facing harm) from a failure mode (how the system went wrong).
  • Recognise AI-specific failure mechanisms, here, fragmented evidence across data sources.
  • Build a control set that survives audit: prevention, detection, and correction.

Educational simulation only. Not a substitute for local clinical safety sign-off or organisational risk assessment.

What you rehearse

Five decisions every hazard entry has to get right.

  1. 01

    Identify the hazard

    A hazard is not a failure mode. Rehearse the distinction on real clinical workflows where the temptation to merge the two is strongest.

  2. 02

    Articulate the clinical consequence

    Vague consequences produce vague controls. Practise describing harm in clinical terms a reviewer cannot argue with.

  3. 03

    Quantify severity and likelihood

    Honest scoring under time pressure, with the reasoning made explicit. The simulator surfaces the assumptions you did not realise you were making.

  4. 04

    Specify a control that controls

    Training is not a control. Policy is not a control. Rehearse the difference on scenarios where the distinction actually matters.

  5. 05

    Decide residual risk

    Accept, monitor, or reject. The simulator asks you to justify the decision and then pressure-tests the justification.

Disclaimer. Independent educational resource — not affiliated with NHS England or any regulatory body. For educational use only.