All modules
Module 04 · 14 minBeginner

AI safety, ethics & responsible use

The guardrails every clinician should know: privacy, HIPAA, bias, transparency, oversight, accountability, explainability, validation, and security.

The nine pillars of responsible clinical AI

None of these stand alone. Strength in one pillar doesn't compensate for weakness in another.

Patient privacy

Protect identifiable patient information at every step — including 'just a quick rewrite.'

HIPAA considerations

Most consumer tools aren't HIPAA-compliant. Use only institution-approved AI for PHI.

Bias

Watch for uneven performance across patient subgroups — bias is a safety issue.

Transparency

Tell patients when AI was meaningfully involved in their care or communication.

Human oversight

A qualified clinician reviews and is accountable for any AI-influenced decision.

Accountability

You sign it, you own it. AI doesn't carry the medical license.

Explainability

Prefer tools that show why they suggest what they suggest, when possible.

Clinical validation

A tool that worked in a study isn't necessarily safe in your specific workflow.

Security

Treat AI integrations as part of your attack surface, not a separate concern.

A simple risk framework: red, yellow, green

When in doubt about an AI use case, place it on this gradient. If it's red, stop. If it's yellow, verify. If it's green, proceed with normal oversight.

Green
Generally safe with normal oversight
  • Approved ambient scribe — clinician reviews and signs the note.
  • Approved tool summarizing a long admission for handoff.
  • Education and reference use, with no PHI.
Yellow
Use with explicit caution and review
  • Drafting a patient message with PHI — only in approved tools, with clinician review.
  • Triage support flagging high-risk patients — always re-evaluated at the bedside.
  • Imaging assistance — interpreted by the responsible radiologist or clinician.
Red
Do not use
  • Pasting PHI into a public consumer chatbot.
  • Acting on AI-generated clinical recommendations without verification.
  • Letting an AI autonomously diagnose, triage, prescribe, or discharge.
Interactive activity

Would you use AI here?

Read each scenario and decide where it falls on the safety spectrum. Choose one of three responses, then rate how confident you were. You'll get an explanation, the safety principle at play, and a running score.

Before you begin

Before using any AI system in healthcare, always ask two questions:

  1. Is this an approved, HIPAA-compliant AI tool?
  2. Is this an appropriate clinical use with the necessary human oversight?
🟢 Appropriate AI Use
🟡 Appropriate With Human Oversight
🔴 Inappropriate / Unsafe
15 scenarios · ~6 min

What would you do?

Pick a response. Each scenario reveals the trade-offs.

What would you do? · Scenario 1

A colleague pastes a full clinic note (with patient name) into a free public chatbot to 'clean it up.' What would you do?

What would you do? · Scenario 2

An AI-generated discharge summary lists a medication the patient never received. You catch it during review. What's the right move?

Reminder: HIPAA isn't optional and "I didn't mean to" isn't a defense. Treat PHI in AI tools the way you'd treat any other disclosure.

Healthcare professional pledge

Principles for responsible AI use

  • I will treat patient privacy as non-negotiable.
  • I will use only AI tools approved by my institution for PHI.
  • I will review every AI-generated word that touches a patient or chart.
  • I will verify clinically meaningful AI output against trusted sources.
  • I will ask about training data, validation, and known limitations before trusting a new tool.
  • I will stay the accountable decision-maker — AI is a tool, not a colleague.
Key takeaways
  • Privacy, oversight, and accountability are the foundation — everything else is technique.
  • Most public chatbots are not HIPAA-compliant. Use only approved tools for PHI.
  • Bias is a patient-safety issue. Ask how a model performs across subgroups.
  • Transparency and explainability help clinicians and patients trust AI appropriately.
  • When in doubt: red, yellow, green. Stop, verify, or proceed with normal oversight.

Safety vocabulary

Flip each card to lock in the definition and a clinical example.

Quick check

Which scenario best fits the 'red' (do not use) tier of the risk framework?

Check your understanding

Five questions on safety, ethics, and clinical accountability.

Knowledge check
Question 1 of 5

You're not signed in — this score won't be saved or counted toward your level average. Sign in first to record it.

A nurse wants to use a free public chatbot to reword discharge instructions that include the patient's name and MRN. The best response is:

Score: 0
Previous: Generative AI