Healthcare | Implementation Guide

    How to Stop a Prior Authorization Agent from Sending More PHI Than the Payer Requested

    Treat over-disclosure as an execution-path problem, not a prompt problem. Convert each payer's documentation requirements into a versioned, machine-enforceable data scope. Enforce that scope on the agent's tool calls and outbound payloads by allowing, redacting, blocking, or escalating each data element. Retain a per-submission record of the scope applied and every decision made.

    1. 1

      Retrieval

      Queries against the EHR, notes, and attachment stores return broad clinical context. Limit what each retrieval can return for the active payer and service.

    2. 2

      Context assembly

      Retrieved content is combined into the model's working context. Anything present here can be copied into an outbound field.

    3. 3

      Tool calls

      Arguments passed to submission and attachment tools can carry free-text narratives, identifiers, and unrelated diagnoses.

    4. 4

      Outbound payload

      The final request and any attached documents are what actually leave the organization. This is the last point of enforcement.

    5. 5

      Allow

      The element is in scope for this payer and service. Release it and record the match.

    6. 6

      Redact

      The element sits inside an otherwise permitted field or document, such as an unrelated diagnosis in a clinical note. Remove it and record what was removed.

    7. 7

      Block

      The element or attachment has no basis in the scope. Stop the call and return the reason to the agent.

    8. 8

      Escalate

      The submission is non-routine, or the policy requires a human decision. Hold it for approval and record the approver and outcome.

    Evidence to retain for each submission

    • Payer request identifier and the scope definition version applied.
    • Owner of the scope definition and the documented basis, such as the transaction exemption or reasonable reliance.
    • Per-element decision: allowed, redacted, blocked, or escalated.
    • Approver identity, time, and outcome for every escalation.
    • A record of what was transmitted, sufficient to compare against the scope.
    • Agent, tool, and timestamp for each tool call.

    What minimum necessary means for an automated submission

    HHS describes the minimum necessary standard as a requirement to take reasonable steps to limit uses, disclosures, and requests of PHI to what is needed for the intended purpose. For routine requests and disclosures, a covered entity may use standard protocols without reviewing each one individually. For non-routine ones, it must develop criteria and apply them case by case. That split maps well to automation: routine prior authorization types can be handled by enforced policy, and unusual ones need a different path.

    Two provisions shape how you define scope. First, a disclosing covered entity may rely, if reasonable under the circumstances, on another covered entity's request as the minimum necessary. A payer's stated documentation requirements can therefore serve as the scope baseline, but only where reliance is reasonable. Second, HHS exempts uses and disclosures required for compliance with applicable transaction standards, including data elements that are required or situationally required in those transactions. Discretionary clinical content outside those elements stays subject to the standard.

    Responsibility does not move to the requester. NCVHS has noted that the custodian of the PHI is responsible for the determination, and that a patient's entire record may at times be the minimum necessary. A narrow policy therefore needs a documented exception path. Confirm with counsel how the exemptions and reliance provision apply to each submission category.

    Define a payer-specific data scope

    Start from the payer's own statement of what it wants. Under CMS-0057-F, impacted payers must maintain a Prior Authorization API that identifies documentation requirements for items and services that require prior authorization, excluding drugs. That requirement is effective 2027-01-01. A 2026 CMS proposed rule would adopt the HL7 FHIR Da Vinci CDex IG for attachments, which CMS says would let plans be explicit about requested data. As of November 2026, that proposal is not final. Both sources are machine-readable inputs from which a scope policy can be derived.

    Express the scope as element-level rules rather than document-level permissions. Each rule should say which data element may appear, in which field or attachment, for which payer and service. Mark each element as required by the transaction standard, requested by the payer, or neither. The third category is the one the enforcement layer must act on. Scope definitions are inferences from payer requirements, so they need review and ownership like any other policy.

    • Inventory payer and service types, and record each payer's documentation requirements per item or service as versioned scope definitions.
    • Classify each element as transaction-required, payer-requested, or out of scope.
    • Classify submission types as routine or non-routine, and document the standard protocol or criteria for each.
    • Bind every submission to the specific payer request that defined its scope.
    • Name an owner for each scope definition and for exception approvals.
    • Update definitions when requirements change, including the 2027-01-01 API date and any final rule on attachments.

    Tradeoffs and operating decisions

    A tight scope can remove clinical content that supports medical necessity. The likely result is a payer response asking for more information, which adds a cycle. The better response is to tune scope definitions with the clinical and compliance owners, not to widen the agent's default access. Where the whole record is genuinely needed, route that submission through the exception path and record the rationale.

    Redaction and blocking trade completeness against risk. Redact when the surrounding content is needed and the extra element is separable. Block when the element has no basis in the scope. Tiered handling keeps review effort proportionate: enforce automatically on routine types and require human approval for non-routine ones.

    Finally, treat scope as living policy. Payer requirements change, the API date is approaching, and a proposed attachment standard may be finalized. Version every definition so that any past submission can be evaluated against the scope in force when it was sent.

    From payer request to evidence

    Payer request

    Documentation requirements define what the submission may contain.

    Scope policy

    Requirements become versioned, element-level rules with a named owner.

    Runtime enforcement

    Tool calls and outbound payloads are allowed, redacted, blocked, or escalated.

    Evidence

    Each submission carries a record of scope and decisions.

    Enforce payer scope where PHI leaves

    See how policy enforcement in the execution path and audit-ready evidence apply to prior authorization agents handling PHI.

    Talk to an Expert
    Next step

    How does your AI governance program compare?

    See where your program has gaps in less than 2 minutes.