Standards › Standard 2D › Policies
Policy · AI-POL-001 · Version 1.1 · Approved by the ASF Expert Committee
Supports: 2D.4
At a glance
- No AI system supports diagnosis, triage or care decisions without an assessment against WHO AI governance guidance (E-011).
- Each AI system has a named clinical owner and a human oversight rule.
- Record each assessment on AI-FRM-001.
- Review every AI system each year.
- Stop and get help straight away if: an AI system is used in clinical decisions with no assessment or no named owner.
1. Purpose
To make sure AI systems used in diagnosis, triage or care decisions are introduced and governed in line with WHO guidance on the ethics and governance of AI for health [E-011].
2. Scope
| Applies to | Clinical leads, IT lead, chief executive, ethics or governing body |
| Locations | All clinical areas at the facility |
| When | Before introduction, at each change, and yearly |
| Does not apply to | Simple non-clinical automation such as scheduling, unless it affects clinical decisions |
3. Definitions
| Term | Meaning |
|---|---|
| AI system | Software that uses machine learning or similar methods to produce outputs such as diagnoses, risk scores or recommendations |
| Human oversight | A named clinician who can review, change or reject an AI output before a care decision |
4. Background
AI can help clinicians, but it can also be wrong in ways that are hard to see. WHO’s 2021 guidance sets principles such as protecting autonomy, promoting well-being and safety, transparency, accountability, equity and sustainability [E-011]. Criterion 2D.4 asks each system to be introduced in line with that guidance.
5. Responsibilities
| Role | Responsibility | Competence |
|---|---|---|
| Chief executive | Approves the policy and high-impact systems | Executive |
| Clinical lead for each AI system | Owns the system and the oversight rule | Clinical |
| IT lead | Assesses data and technical risk | Health informatics (EC) |
| Ethics or governing body | Reviews high-impact systems | Ethics and governance (EC) |
6. Materials and equipment
| Item | Requirement | Condition |
|---|---|---|
| AI system assessment (AI-FRM-001) | Current | One per system |
| WHO guidance (E-011) | Current copy | Available to leads |
7. Safety
- AI outputs are advice, not orders. A clinician makes and records the final decision.
- Do not use an AI system outside the population or setting it was validated for.
8. Before you start
- Check the system is on the register of AI systems.
9. Procedure
Step 1. Register the system.
- List each AI system in use: purpose, vendor, version, data used, and the clinical owner.
Key point: An unregistered system cannot be governed.
Step 2. Assess against WHO guidance.
- Complete AI-FRM-001 for each system against the WHO principles: autonomy, safety, transparency, accountability, equity and sustainability.
- Record the evidence for each principle.
Key point: The assessment must show how each principle is met, not only that it is claimed.
Step 3. Set human oversight.
- Write who reviews each output, when, and what they may do with it.
- Record the rule on AI-FRM-001.
Key point: Oversight must be real. A rubber stamp is not oversight.
Step 4. Decide and introduce.
- The approver decides on the basis of the assessment.
- Introduce with training and monitoring (AI-SOP-001).
Key point: The decision and its reasons are the evidence.
Step 5. Review yearly.
- Review each system yearly against the assessment and the monitoring results.
Key point: Models and data change over time.
10. Decision points
- If the assessment cannot show equity or safety for the patient group, the system is not used for that group.
11. Special situations
| Situation | Action |
|---|---|
| Vendor update changes the model | Repeat the assessment before the update is used clinically. |
12. Escalation
- Any AI-related safety concern goes to the chief executive and INC-POL-001 at once.
13. Monitoring
| Measure | Target | Frequency | Owner | If not met |
|---|---|---|---|---|
| AI systems registered with assessment | 100% | Yearly | IT lead | Stop use until assessed |
| Named oversight rule per system | 100% | Yearly | Clinical leads | Escalate |
| Assessments reviewed yearly | 100% | Yearly | Chief executive | Escalate to governing body |
14. Records
| Record | Kept in | Retention |
|---|---|---|
| AI system register and AI-FRM-001 assessments | Governance file | the period set by the facility’s records policy years after the system is retired |
15. Training
- Clinical users are trained on each AI system and its oversight rule before use (EC).
16. Related documents
| ID | Title | Use |
|---|---|---|
| AI-FRM-001 | AI System Assessment | Used in step 2 |
| AI-SOP-001 | AI Performance and Safety Monitoring | Monitoring after introduction |
| SOP-N28 | Responsible Use of AI Policy (generic) | Basis |
| DIG-SOP-002 | Digital Care System Assessment | Overall digital assessment |
17. Evidence
| Reference | Note |
|---|---|
| E-011 | WHO (2021), Ethics and governance of artificial intelligence for health. Used as the basis for AI governance. Link and wording to verify before publication. |
| E-009 | ASF Expert Committee consensus. Assessment criteria and high-impact thresholds to be set by the committee. |
18. Change history
| Version | Date | What changed | Approval |
|---|---|---|---|
| 1.0 | the date of approval | First version | Approved by the ASF Expert Committee |
| 1.1 | the date of approval | Internal quality review: corrected cross-references, status wording and document control | Approved by the ASF Expert Committee |
E-learning for this document
- Artificial Intelligence in Healthcare Safety (AI-001)
- Artificial Intelligence Systems for Care (TELE-10)
References
- World Health Organization (2021). Ethics and governance of artificial intelligence for health: WHO guidance. Geneva: WHO. Open source (E-011)
- Accréditation Sans Frontières (2026). ASF Expert Committee consensus on Standard 1. (E-009)