Guardrails for nurses

Who it's for

Everyone — ten rules of thumb for caring, teaching, learning or building with any AI tool.

Red line

Protect identity — never enter patient, family or staff identifiers into a tool not approved for it.

Nurse checks

A human decides — verify against NICE, the BNF and local policy, and look for what and who is missing.

These ten guardrails are the toolkit's own rules of thumb. They apply whatever the tool, and whether you are caring, teaching, learning or building.

Before you prompt

1–3. Purpose, Approval & Privacy

• 1. Purpose first: Is AI the right tool?
• 2. Approved tools only: Use institutional sign-in.
• 3. Protect identity: Never enter patient or staff identifiers.

While reviewing output

4–7. Human Judgement & Equity

• 4. A human decides: Registered nurses own decisions.
• 5. Check before use: Verify against NICE / BNF.
• 6–7. Check what & who is missing: Spot omissions and bias.

Professional accountability

8–10. Candour, Competence & Safety

• 8. Be open: Disclose AI use honestly.
• 9. Within competence: Only use what you can verify.
• 10. Report & learn: Escalate AI errors locally.

The ten guardrails

1. Purpose first. Be clear about what you are trying to achieve, and whether AI is the right way to achieve it.
2. Approved tools only. Use tools your organisation or university has approved, for the purposes they're approved for.
3. Protect identity. Don't enter identifiable information about patients, families or colleagues into a tool that isn't approved for it. Anonymising isn't always enough when details are rare or combined.
4. A human decides. AI can suggest. A registered professional makes and owns every decision about care.
5. Check before use. Verify clinical content against authoritative sources, such as NICE, the BNF and local policy, before acting on it, teaching it or saving it to a record.
6. Look for what is missing. Check for omissions as carefully as for errors.
7. Look for who is missing. Ask whether the output could disadvantage people because of ethnicity, disability, age, gender, sexuality, language or class.
8. Be open. Tell people when AI has helped produce something they rely on, following your organisation's policy, and disclose AI use in academic work.
9. Stay within your competence. Don't use AI to do things you couldn't do, explain or check yourself. See Novice to expert.
10. Report and learn. Report problems with AI tools through local incident systems, and share what you learn.

Mapping to the NMC Code

Guardrail	Closest Code standards
1, 4, 9	13 (work within the limits of your competence), 19 (reduce potential for harm)
2, 3	5 (respect privacy and confidentiality)
5, 6	6 (practise in line with the best available evidence), 10 (keep clear and accurate records)
7	1 (treat people as individuals and uphold their dignity) and 20 (treat people fairly and without discrimination)
8	20 (act with honesty and integrity)
10	16 and 19 (act without delay if you believe there is a risk to safety; reduce potential for harm)

Read the full Code: NMC Code online.

Wider guidance

• UK government: the AI Playbook for the UK Government sets out principles for public sector use of AI.
• NHS England: guidance on AI-enabled ambient scribing products.
• Regulation of AI products: the AI and Digital Regulations Service explains the rules for health and care.

---

Quick self-check: The 10 NCD Guardrails

Quick Clinical Self-Check

Why do Guardrails 6 and 7 ask nurses to check both 'what is missing' and 'who is missing' when reviewing an AI output?