Free governance template · MSoP v1.0 · CC BY 4.0

The Model Scope of Practice

A hospital would never let a clinician practice without defined privileges: which patients, which procedures, which settings, and where the boundaries are. Yet AI models are routinely deployed on an IT ticket, with no declared scope at all. The Model Scope of Practice fixes that with one page. It adapts the logic of medical-staff bylaws to software: what this system is authorized to do, for whom, and the exact conditions under which it must stop.

Why a scope, not a model card

A model card describes what a model is. A scope of practice declares what it is permitted to do here, in this institution, on these patients, and it makes the boundary operational: an input outside the declared scope, a lab code that changed in an EHR upgrade, a population the model was never validated on, becomes a rejectable event rather than a silent inference. The scope is the credentialing stage of Closed-Loop Defensibility: nothing reaches production without one.

The template

Copy it whole, or hand it to your medical-staff office and counsel to adapt. It is CC BY 4.0, so it may be used in any institution without asking.

MODEL SCOPE OF PRACTICE (MSoP) CHARTER — TEMPLATE v1.0

SYSTEM
  Name / version:            __________________________________
  Vendor (or "in-house"):    __________________________________
  FDA status:                [ ] 510(k) K-number: ______  [ ] Not FDA-regulated  [ ] Other: ______
  Governance record (MGR):   URL or "none yet": ___________________

ACCOUNTABLE OWNERS (all required before production)
  Clinical owner (CMO / dept chair): _____________________  Date: ______
  Informatics owner:                 _____________________  Date: ______
  Assigned proctor (named human):    _____________________  NPI: ______

AUTHORIZED SCOPE
  Patient populations authorized:    __________________________________
  Encounter types authorized:        __________________________________
  Clinical units / settings:         __________________________________
  Permitted inputs (data types):     __________________________________
  Permitted outputs / actions:       __________________________________

BOUNDARIES (out of scope — the system must not be used for these)
  Excluded populations:              __________________________________
  Contraindicated encounter types:   __________________________________
  Inputs that must be rejected:      __________________________________
  Actions the system may never take: __________________________________

PRIVILEGING TIER (see Autonomous Agent Privileging)
  [ ] Provisional  — human co-signature on every output; read-only; shadow runs.
  [ ] Conditional  — sampled human audit at a stated rate; limited/draft write access.
  [ ] Full         — autonomous action within scope; continuous surveillance.

RUNTIME CONDITIONS (the loop, per CLD-SPEC-v1.0)
  Drift / disparity / override are measured on this cadence:  ______________
  A threshold breach automatically changes privileges to:     ______________
  On breach, this named human is alerted:                     ______________
  Target Governance Decision Latency (breach to human decision): ____ hours

REVIEW
  This charter is reviewed at least every:  ______  and on any model version
  change, retraining event, or upstream input change (e.g. a lab code or FHIR
  resource change) affecting the inputs above.

  Signed: ____________________ (Clinical)  ____________________ (Informatics)
  Effective date: ____________   Charter version: ____________

Licensed CC BY 4.0 by The Healthcare AI Institute. A governance template for
adaptation by the organization and its counsel; not legal, clinical, or
regulatory advice.

What to do with it once it is signed

The charter is the first of five stages. Its boundaries become the inputs the runtime must reject; its privileging tier becomes what the circuit breaker can change; and the whole charter, with its owners and version, belongs in the system’s Model Governance Record so a surveyor or a court can see it existed, dated, before the event they are asking about.

Requiring vendors to arrive with a scope already declared? The Model Governance Rider puts it in the contract. Assessing where your own program stands across all five focus areas? The readiness score produces a board-ready gap report in ten minutes.

MSoP v1.0 is part of CLD-SPEC-v1.0, published by The Healthcare AI Institute under CC BY 4.0. It is a governance template for adaptation by the organization and its counsel, not legal, clinical, or regulatory advice.