A risky AI use alert gets an owner
The sample response names who reviews and who may act.
Coming
See every demonstrationHEALTHCARE | SAMPLE WORKFLOW
Review access before clinical operations use an assistant.
Show me the test, not the slide that says it was tested. A fictional provider runs an administrative workflow on sample data, checks identity and data handling, exercises the escalation route, and keeps clinical decisions with qualified staff.
01 Source02 Boundary03 Human decision04 Evidence
HOW IT WORKS
The control path in practice.
- Map the administrative task and approved sample data.
- Check user and agent access to each source.
- Test failure cases and document the escalation route.
- Review operating evidence before any production use.
THE WOUND
How do you prove the controls were tested, not just presented?
With a test you can show: identity and data handling checked on sample data, the escalation route exercised, and the operating evidence reviewed before any production use.
You have been handed the slide that says it is tested. This page is about the test itself, run on sample data in your Microsoft tenant, with clinical decisions kept with qualified staff throughout.
THE MICROSOFT INSIGHT
What security capability is already in your E5 license and not switched on?
Microsoft Security Copilot, which is included for eligible Microsoft 365 E5 and E7 customers and has been rolling out in phases since November 18, 2025.
Microsoft Security Copilot is included for eligible Microsoft 365 E5 and E7 customers; the rollout started November 18, 2025 and continues in phases. (Microsoft Learn, checked Oct 31, 2026)
Activate what the license includes and the budget line is operating effort. Buy beside it and the budget line is paid twice.
THE SAMPLE WORKFLOW
How does an administrative AI workflow earn its place in a provider?
By being mapped, access-checked, failure-tested and evidence-reviewed on sample data before a single real record is touched.
- Map the administrative task and the approved sample data.
- Check user and agent access to each source.
- Test the failure cases and document the escalation route.
- Review the operating evidence before any production use.
THE EVIDENCE
What evidence does HIPAA work need from the AI estate?
Who had access, what the agent did, who approved it and what the policies caught, filed in the format your compliance officer already uses.
Microsoft Purview Insider Risk Management ships a Risky AI usage policy template that scores prompts and responses containing sensitive information in Microsoft 365 Copilot and agents, and browsing to generative AI sites. (Microsoft Learn, checked Oct 31, 2026)
With the policy and the record, a risky prompt is an alert with an owner. Without them, it is a breach assessment.
THE FIRST RUNG
Where does a stretched security lead start?
With the free AI security check, then the Agent 365 Identity and Security Assessment, then managed security under an authority you set in writing.
The authority matrix says who may disable an account at 2 a.m. and who gets told. It is written with you and your compliance officer before the service starts, and it is part of the evidence pack.
AFTER TWO WEEKS
What will be working after two weeks?
- The administrative workflow and its sample data are mapped and agreed.
- Access for users and agents to each source has been checked and the gaps named.
- The failure cases have been run and the escalation route is written down.
- You hold the production review scope and its quote as a written decision.
HONEST ABOUT LIMITS
What this does not do yet
It does not rely on a preview feature for the SIEM layer. ISOC in Microsoft Defender, in preview, brings SIEM capabilities into the Defender portal for eligible Microsoft 365 E5 and E7 customers without an active Microsoft Sentinel workspace, with 30 days of included retention for Defender data during this preview phase. (Microsoft Learn, checked Oct 31, 2026, preview) The plan uses the controls that are fully released today and records the preview as a candidate for the next review.
It does not touch a clinical decision: those stay with qualified staff, and the workflow is administrative by design.
It does not make a HIPAA attestation for you: the evidence is filed, your compliance officer owns the statement.
THE ONE THING TO REMEMBER
Show the test, not the slide that says it was tested.
Said another way: sample data first, real records only after the evidence review.
Said a third way: evidence, not exhaustion.
SHOW YOUR WORK
See the control path on a fictional bank.
DIRECT ANSWERS
Questions before you decide.
Does the AI touch clinical decisions?
No. The workflows in scope are administrative, and clinical decisions stay with qualified staff throughout. The permission row makes that explicit.
What is tested before production?
Identity and data handling on sample data, the failure cases, the escalation route and the operating evidence, with your compliance officer reviewing the record.
What is in our E5 license that we are not using?
Often Microsoft Security Copilot, included for eligible E5 and E7 customers, and the Purview policies that produce the evidence HIPAA work needs. The assessment lists what is switched on and what is not.
Who acts at 2 a.m.?
Whoever the authority matrix names, for the actions it grants. Anything outside it waits for a named person, and every action is filed in the evidence pack.
How is it priced?
Quoted after a scoping call. The assessment is scoped separately from the managed service, and the service starts only after the authority matrix is signed.
NEXT STEP
Start with a boundary you can explain.
Choose a focused check, or speak with the team about the workflow you need to improve.