OUT · Output Integrity & Supply Chain
OUT-16Fallback model assurance
Any alternate model an agent may run on is held to the same pre-deployment assurance as the primary. An agent does not silently operate on an unassessed model.
A fallback model is not a free pass. Any alternate model an agent may run on, whether selected deliberately or invoked automatically by a provider's fallback mechanism, is held to the same pre-deployment assurance as the primary: the capability evaluation suite, the versioned evaluation set, adversarial assessment, and a captured behavior baseline. Where a provider may substitute models without notice, the substitution is either contractually constrained, detected and contained at runtime (MON-12), or the fallback is pre-assessed so the substitution lands on assured ground. An unassessed substitution is a material change and is handled as an incident.
Table OUT-16.1 · Maturity criteria · v.1.3
L1 to L5 · the lowest level in scope caps the agentControls exist informally; documentation is incomplete; no consistent review.
Controls are written down; ownership is assigned; review cadence is defined.
Controls are followed in practice; evidence is collected; deviations are tracked.
Every pre-identified fallback model carries a current assessment to the same standard as the primary; provider substitution paths are contractually constrained, detected at runtime, or pre-assessed; no production traffic is served by an unassessed model.
Control effectiveness is measured against documented thresholds; trends are reported.
Controls evolve based on incident learning; maturity gains are evidenced over time.
The v1.3 edition publishes a control-specific criterion at L3 (Operated). The other levels apply the general ladder to this control; see the maturity ladder.
Evidence an assessor asks for
- E.1
Every pre-identified fallback model carries a current assessment to the same standard as the primary.
- E.2
Provider substitution paths are contractually constrained, detected at runtime, or pre-assessed.
- E.3
No production traffic is served by an unassessed model.
Table OUT-16.2 · Framework mappings
5 of 7 frameworks carry a control-level referenceMappings are directional: meeting this control evidences but does not by itself satisfy the external requirement. A middle dot means no control-level reference is published at this edition; see the crosswalk.
Related controls and terms
Cite this control
Human Intelligence. HI-AAF v1.3, OUT-16 Fallback model assurance. Public review edition, July 2026. https://humanintel.net/standard/out/out-16
Edition v.1.3 · July 2026 · Public review · Markdown · PDF
https://humanintel.net/standard/out/out-16
HI-AAF v1.3 is an independent framework published by Human Intelligence. It is not affiliated with NIST, ISO, AICPA, or any government or accreditation body. The framework is offered openly for review during the current draft cycle.