This case study is a composite based on patterns we see across regional health system deployments, illustrating a typical HIPAA-aligned rollout. Specific figures are illustrative.

The challenge

A regional health network operating 14 clinics and a 220-bed hospital wanted to give clinical staff and administrative teams access to AI for documentation support, patient communication drafting, and internal policy lookup — without creating a new avenue for PHI exposure. Their compliance team's requirement was specific: any AI tool touching patient data needed a signed Business Associate Agreement, PHI redaction before data reached a third-party model, and an audit trail detailed enough to support a HIPAA compliance review on request.

Two prior AI pilots had stalled at the security review stage because neither tool could demonstrate consistent PHI handling across every feature, not just the primary chat interface.

The approach

Redaction before generation, not after

Patient identifiers, medical record numbers, and other PHI fields are detected and tokenized before a request reaches any model provider, and re-inserted only in the final response shown to the authorized clinician — meaning no external model provider ever receives raw PHI, regardless of which model is handling the request.

Connector access scoped by role

Clinical staff, billing staff, and administrative staff each see a different slice of the connected knowledge base, enforced by the same role definitions already used in their identity provider — a nurse's AI assistant can retrieve clinical documentation templates, but not billing or HR policy documents, without a separate access system to maintain.

Human approval on anything that becomes part of the medical record

AI-drafted clinical documentation is never filed automatically. A named clinician reviews and approves the draft before it becomes part of the patient record — the AI accelerates the drafting, but the human remains the author of record.

The outcome

0

Instances of raw PHI reaching an external model provider, confirmed via audit log review

14

Clinics onboarded in the initial rollout

100%

Of AI-assisted clinical documentation reviewed by a named clinician before filing

The compliance review that had stalled two previous AI pilots was completed in a single pass, largely because redaction and audit logging were built into the Governed AI Workspace's default configuration rather than something the health network's own engineering team had to assemble and validate themselves.