AI Healthcare Safety Checklist
AI Healthcare Safety Checklist
This checklist helps healthcare teams review AI-supported workflows before they are used in documentation, administration, patient communication drafts, staff training, operational planning, or clinical-adjacent support work.
AI should be treated as draft support until it has been reviewed for privacy, accuracy, source grounding, policy fit, escalation needs, and clinical safety boundaries.
Before Using AI in a Healthcare Workflow
- Confirm the workflow owner and reviewer
- Define whether the workflow is administrative, patient-facing, clinical-adjacent, or high-risk
- Use an approved AI tool for sensitive work
- Remove unnecessary PHI and sensitive details
- Ground output in approved source material
- Define what must be checked before use
- Document edits, approvals, and final decisions
Healthcare AI Review Questions
- Could this output affect patient understanding, care, safety, records, billing, or compliance?
- Does the workflow involve patient identifiers or sensitive information?
- Is the AI output based on approved sources?
- Could the output contain invented facts, unsupported advice, or missing context?
- Who is responsible for final approval?
- What should trigger escalation?
- How will review and final use be tracked?
Do Not Use AI Without Extra Review When
- The workflow involves diagnosis, treatment, prescribing, triage, or urgent symptoms
- The output will be patient-facing
- The workflow includes PHI or confidential records
- The output affects documentation, billing, claims, or compliance
- The situation is sensitive, unclear, high-risk, or outside normal policy