Healthcare AI vendor risk · A ReMyll product
Sample assessment Contact
AITPRM Health™A ReMyll product
Worked scenarios

Worked scenarios: how the review runs.

Three composite accounts drawn from customer programs. Organization names and figures are illustrative; the sequence of events is representative.

The RecordWorked scenarios2026 AI Renewal GuideFAQSample assessmentTrust Center
01Regional health system · 6 hospitals

An ambient documentation vendor, reviewed before signature

The privacy officer had a security questionnaire and a redlined BAA, and no way to ask about training rights or the audio consent model. The scoped set surfaced an unnamed foundation model provider in the subprocessor register.

Roles involved · Privacy, Compliance, Legal
01Scoped set of 118 questions issued to the vendor; answered in 9 days
02Tier-4 finding on subprocessor disclosure; review paused pending resolution
03Suggested clause template reviewed by counsel; provider named and bound
04Committee approved with conditions; record retained for 2027 renewal
FindingF-07 · Tier 3 · D2

Foundation model provider not named in the BAA or any flow-down agreement. Vendor answer to Q 02.3 identifies “a leading commercial LLM provider” without naming it; the architecture diagram (E-11) names it. Evidence rated as a documentation gap.

Sample contract control · CC-07For counsel

Vendor shall identify in Schedule B each third party that provides, hosts or operates any machine learning model used to process Customer Data, and shall deliver executed flow-down agreements within thirty (30) days of the Effective Date.

Illustrative. Scenario, finding and sample clause templates are fictitious. Not legal advice.

02Behavioral health provider · 42 CFR Part 2

A predictive risk tool already running, brought under review at renewal

The tool had been in production for two years without an AI-specific review. The renewal was the first opportunity to set terms on derivative data and model incident notification.

Roles involved · Compliance, Clinical leadership
01Reviewed as a renewal: prior contract loaded as baseline
02Part 2 overlay added 14 questions on redisclosure and consent
03Tier-3 finding: no notification duty for model drift; control added at renewal
04Reassessment scheduled at 12 months with version comparison
FindingF-04 · Tier 3 · D7

Session audio retained for 90 days for “quality purposes” with no Part 2 consent language and no patient-facing disclosure. Evidence: retention policy verified; consent model missing.

Sample contract control · CC-04For counsel

Vendor shall not retain audio or transcript content beyond completion of the encounter note, shall obtain and document consent in the form Customer specifies, and shall not redisclose Part 2 records except as 42 CFR Part 2 permits.

Illustrative. Scenario, finding and sample clause templates are fictitious. Not legal advice.

03Medicare Advantage plan

Utilization management vendor, aligned with CMS expectations

The plan needed to show that coverage determinations reflected individualized review, not algorithmic output alone, and needed the evidence to be inspectable.

Roles involved · Compliance, Procurement, Medical management
01D4 human-in-the-loop questions answered with workflow evidence
02Evidence rated stale on one SOC 2 report; refreshed before decision
03Board packet exported for the compliance committee
04Portfolio view now covers 11 AI vendors on one rubric
FindingF-02 · Tier 4 · D4

Prior-authorization recommendations can be batch-approved by a reviewer without opening the individual case. No individualized determination is evidenced. Review paused pending resolution.

Sample contract control · CC-02For counsel

Vendor shall configure the Service so that no coverage determination is finalized without a named clinical reviewer opening and attesting to the individual case, and shall log each attestation with reviewer identity and timestamp.

Illustrative. Scenario, finding and sample clause templates are fictitious. Not legal advice.

Next step

Run the same framework on your vendor.

The findings in these cases came from specific questions. On the demo you see those questions scoped against your own deployment.