HIPAA.technology — AI Learning Lab

An agent-native learning experience for physicians and medical-practice leaders, built for the OpenAI WebMCP Challenge.

Launch the live experience

Doctors choose a realistic AI use case, explore how information moves through it, and collaborate with their browser agent to build a safe 21-day adoption mission. The agent can inspect the current learning context, compare scenarios, focus the shared interface, prepare a risk briefing, and stage a plan. Consequential decisions remain visibly human-owned.

Its signature Practice Website Debrief lets a doctor enter a public domain. The agent reviews the public site, then writes a constructive debrief back into the shared experience: what is already strong, what could be even better, and the safest practical AI next move. It is an educational public-site review, never a HIPAA compliance audit.

The experience is informed by ChatGPT Deep Research, Middleware Encryption for HIPAA-Compliant Use of Frontier LLMs in U.S. Healthcare. Its central teaching is deliberately sharper than “encrypt the prompt”: teams must map the entire information flow, enforce minimum-necessary disclosure, validate every covered feature and vendor relationship, govern logs and agent tools, and preserve human accountability.

Why WebMCP

Ordinary healthcare education pages give every visitor the same article and force agents to infer meaning from pixels. This app exposes a structured, state-aware vocabulary for the learning journey. WebMCP lets the agent understand the selected scenario, operate the same interface the doctor sees, and turn education into an accountable next step.

The app registers tools directly with document.modelContext.registerTool():

• inspect_adoption_workspace — reads the current scenario and mission state
• focus_learning_lesson — focuses one HIPAA + AI learning principle
• select_clinical_ai_use_case — changes the shared visual focus
• assess_use_case_readiness — returns a structured educational readiness brief
• stage_21_day_adoption_mission — stages a mission in the visible UI
• revise_adoption_mission — revises a draft and resets human approval
• prepare_practice_website_review — stages a public-domain review mission and rubric
• deliver_practice_website_debrief — writes the agent's sourced findings into the shared UI
• set_contextual_guidance — updates the floating AI Guide for the current section
• focus_mission_week — aligns the visible plan with the agent's current focus

Human-agent authority boundary

The agent may inspect, explain, compare, focus, and draft. A human must approve the mission, accept policy guardrails, and make every clinical or compliance decision. This is a deliberate product decision, not an API limitation.

Run locally

Requires Node.js 22.13 or newer.

npm install
npm run dev

To test with ChatGPT site tools:

1. Update to the latest ChatGPT desktop app.
2. Use a Work or Codex conversation with GPT-5.6 Sol or GPT-5.6 Terra. GPT-5.6 Luna currently has WebMCP disabled, and site tools are not currently available in Enterprise or Edu workspaces.
3. Turn on Settings → Browser → Permissions → Enable site tools.
4. Open the built-in browser from that same conversation (browser button or Ctrl+Shift+B on Windows), then navigate to https://lab.hipaa.technology.
5. Select Site tools in the browser address bar, open Available site tools, and confirm the ten tools are listed.
6. Keep the page open and send the demo prompt from the site in that same conversation's message box.

For manual testing in Chrome 149+, enable chrome://flags/#enable-webmcp-testing and chrome://flags/#devtools-webmcp-support, then relaunch Chrome.

In Chrome DevTools, open Application → WebMCP to inspect the registered schemas, manually execute tools with parameters, and review invocation results.

Build

npm run build

Safety

This is an educational demonstration using synthetic scenarios. It accepts no patient data and is not legal advice, compliance certification, or clinical guidance.

License

MIT