Skip to main content

Responsible Use

GOV.UK AI Guardrails — Adapted for Nursing

Source

This page adapts the guardrails published by the Government Digital Service's AI Engineering Lab (gds-dtx/aiengineeringlab, Open Government Licence v3.0). The Lab's material is an alpha resource for government departments adopting AI coding assistants; the guardrails were last updated in July 2026.

It is not NHS or NMC policy. We use it because its structure (what data never to share, where AI may and may not be used, how to check its output) translates well to nursing. In practice, follow your organisation's information governance policy, the NMC Code and UK GDPR. For wider government guidance on AI, see the AI Playbook for the UK Government.

Why This Matters for Nursing​

The AI Engineering Lab guardrails give government technology teams a clear, practical structure for using AI safely. As nurses increasingly use AI in education, documentation and clinical decision support, the same structure offers a useful way to teach safe use, even though the guardrails were written for software teams rather than healthcare.

This page translates the government's technical guardrails into nursing-specific language and scenarios, connecting them to the AI Nursing Constitution and NMC professional standards.


1. Data Handling — What You Must Never Share with AI​

GOV.UK Reference: G-DH-01, G-DH-02

The guardrails limit what data can be shared with AI tools according to its classification and prohibit certain types outright. For nursing, this means:

❌ Never Share with AINursing Examples
Patient-identifiable data (PII)Names, NHS numbers, dates of birth, addresses
Health informationMedical records, diagnoses, test results, care plans with patient details
Authentication credentialsNHS login passwords, smartcard PINs, API keys
Biometric dataPatient photographs with identifiable features
Case detailsSafeguarding referrals, incident reports with names
Clinical Scenario

Wrong: Pasting a patient's discharge summary into ChatGPT and asking it to "write a care plan."

Right: Writing a de-identified clinical scenario (no names, no NHS numbers, no dates of birth) and asking for care plan structure guidance.

Prompt Hygiene for Nurses​

Before entering any clinical text into an AI tool:

  1. Remove all identifiers — names, NHS numbers, DOB, addresses
  2. Generalise the scenario — "A 72-year-old female" not "Mrs Smith"
  3. Never paste directly from clinical systems — always rewrite in your own words
  4. Start a new chat when switching between patient scenarios

2. Usage Boundaries — Where AI Should and Shouldn't Be Used​

GOV.UK Reference: G-UB-01, G-UB-02

Appropriate Uses in Nursing Education​

Use CaseNotes
Learning about clinical conditionsUse AI to explain pathophysiology, pharmacology
Generating practice scenariosCreate de-identified case studies for teaching
Exploring care planning frameworksAsk AI about ADPIE, SBAR, ABCDE structures
Drafting reflective writingUse as a starting point, then personalise
Understanding research papersSummarise evidence-based practice articles
Exploring NMC proficiency standardsQuery what competencies relate to a topic

Prohibited Uses​

Use CaseRationale
Making autonomous clinical decisionsAI cannot replace registered nurse judgment
Submitting AI-generated work as your ownAcademic integrity / NMC professional standards
Sharing patient data with AI toolsData protection, Caldicott principles
Using AI for safeguarding decisionsRequires qualified human professional assessment
Bypassing clinical protocolsAI must support, not circumvent, safety processes

3. Ethical Use — Human Judgment Remains Essential​

GOV.UK Reference: G-ET-01, G-ET-02, G-ET-03

The guardrails draw a clear line: AI may assist with research and drafting, but humans must decide on safety-critical matters (G-ET-03). This mirrors the AI Nursing Constitution's core principle.

Clinical Safety Review Triggers​

The guardrails require a clinical safety review, owned by a Clinical Safety Officer, for healthcare and clinical systems (G-ET-01). Adapted for nursing education, that suggests:

System TypeReview RequiredWho Reviews
AI tools used in clinical placementsYes — clinical safety assessmentPractice supervisor / assessor
AI-assisted care planningYes — nursing judgment verificationRegistered Nurse
AI-generated patient education materialsYes — accuracy reviewClinical educator
AI tools processing student performance dataYes — GDPR assessmentData Protection Officer
AI in simulation/OSCE scenariosYes — educational validity checkProgramme lead

The "AI Didn't Write This" Rule​

The guardrails require AI use to be documented and transparent (G-ET-02). For nursing students:

  • Always declare when AI has been used in assessed work
  • Understand what the AI generated — you must be able to explain it
  • Verify clinical accuracy against authoritative sources (BNF, NICE, NMC)
  • Take professional responsibility for any AI-assisted output you submit or act on
NMC Alignment

Under the NMC Code you are responsible for the records you make (standard 10), for working within your competence (standard 13) and for reducing the potential for harm (standard 19). That applies equally to AI-assisted work. If you use AI to help write a care plan and it hallucinates an incorrect drug dose, you are accountable for what you use, not the AI.


4. Agentic AI — When AI Acts Autonomously​

GOV.UK Reference: G-AG-01 to G-AG-08

The guardrails classify AI coding tools by their autonomy level (G-AG-01). The nursing examples below are our adaptation:

LevelDescriptionNursing ExampleControl Required
L1 — SuggestiveShows suggestions you accept/rejectAutocomplete in clinical documentationMinimal
L2 — AssistiveGenerates content you reviewAI drafts SBAR handover, you reviewReview before use
L3 — CollaborativeMulti-step with checkpointsAI generates a full lesson planCheckpoint reviews
L4 — AutonomousExtended operationAI monitors patient deterioration alertsContinuous oversight
L5 — Fully AutonomousHours/days without human inputNot appropriate for clinical settingsN/A
caution

For nursing and healthcare contexts, L4 and L5 autonomy levels require explicit clinical governance approval before deployment. Most nursing AI use cases should operate at L1–L3 with human-in-the-loop review.

Kill Switch Principle​

The guardrails require an accessible, tested kill switch (a way to stop the AI immediately) for the most autonomous tools, L4 and L5, and recommend one at L3 (G-AG-03). In nursing terms:

  • You must always be able to override AI recommendations
  • AI must never lock you out of manual clinical decision-making
  • If an AI system fails, you must be able to deliver care without it

5. Output Validation — Never Trust, Always Verify​

GOV.UK Reference: G-OV-01 to G-OV-04

The VERIFY Framework for Nursing AI Output​

StepActionNursing Application
V — Verify factsCheck clinical accuracyCross-reference with BNF, NICE guidelines
E — Examine biasCheck for demographic biasDoes the output account for diverse skin tones, ages, cultures?
R — Review logicTrace the reasoningDoes the care plan follow ADPIE? Is the triage category logical?
I — Identify hallucinationsSpot invented informationDoes that drug dose exist? Is that NICE guideline real?
F — Flag uncertaintyNote where you're unsureEscalate to a supervisor or mentor
Y — Your judgmentApply professional reasoningDoes this align with your clinical experience?
Hallucination Risk

AI models can confidently state incorrect information, including invented guideline numbers, drug doses and references that look real. Always verify guideline references against the original source, such as the NICE website or the BNF.


6. Self-Assessment Checklist​

Before using an AI tool in your nursing education or practice, work through this checklist, adapted from the AI Engineering Lab guardrails:

Pre-Use Checklist​

  • I understand what data I can and cannot share with this AI tool
  • I have removed all patient-identifiable information from my prompt
  • I know the AI tool's limitations and potential for error
  • I have an authoritative source to verify the AI's output against
  • I understand my professional accountability for any AI-assisted output
  • I am using the AI to support (not replace) my learning and clinical judgment
  • I will declare AI use where required by my institution's academic integrity policy
  • I have a plan for what to do if the AI gives incorrect or harmful output

Post-Use Checklist​

  • I have verified the clinical accuracy of the AI's output
  • I have checked for potential bias (age, gender, ethnicity, skin tone)
  • I can explain the AI's output in my own words
  • I have documented my AI use where required
  • I would be comfortable defending this output to a practice assessor

How This Connects to Our AI Nursing Constitution​

The GOV.UK AI Engineering Lab guardrails and our AI Nursing Constitution share the same foundational principles:

GOV.UK GuardrailAI Nursing Constitution Equivalent
G-ET-03: Human judgment requirementsThe Nurse Decides — AI advises, the nurse acts
G-DH-02: Prohibited data typesPatient Privacy — never share identifiable data
G-ET-02: Transparency and documentationTransparency — always declare AI use
G-AG-07: Meaningful human reviewProfessional Accountability — understand what the AI generated
G-UB-01: Prohibited use casesSafety First — AI must never make autonomous clinical decisions
G-OV-01: Factual verificationEvidence-Based — verify against authoritative sources

The AI Nursing Constitution works as a nursing-specific extension of these guardrails, similar to the department-specific guardrails the Lab encourages organisations to create.


Further Reading​