# OUT-15 Model availability & continuity

Domain: OUT Output Integrity & Supply Chain · HI-AAF v1.3 · Public review · https://humanintel.net/standard/out/out-15

Model availability is treated as a supply-chain risk: each agent documents a continuity posture, fallback path, trigger conditions, and permitted degraded behavior, for loss of its primary foundation model.

Model availability is treated as a supply-chain risk, not an assumption. For each agent, the organization documents what happens when the primary foundation model becomes unavailable, whether through provider outage, version deprecation or sunset, or a regulatory or government directive that withdraws access. The continuity posture specifies the fallback path (an alternate model assured under OUT-16, graceful degradation, or a safe-stop that invokes the kill switch under ACT-07), the conditions that trigger it, and the behavior the agent is permitted to exhibit while degraded. Loss of model is entered in the Agent Risk Register (GOV-03), the posture is recorded in the Behavior Charter (GOV-02), and it is exercised during incident-response drills (MON-07).

## Maturity

- L1 Ad hoc: Controls exist informally; documentation is incomplete; no consistent review.
- L2 Documented: Controls are written down; ownership is assigned; review cadence is defined.
- L3 Operated: Controls are followed in practice; evidence is collected; deviations are tracked. Published criterion for OUT-15: A documented continuity posture exists for every production agent; a failover or safe-stop has been exercised in a drill or live event within the documented cadence: at least annually and on material change; loss of model is represented in the Agent Risk Register.
- L4 Measured: Control effectiveness is measured against documented thresholds; trends are reported.
- L5 Continuously Improved: Controls evolve based on incident learning; maturity gains are evidenced over time.

## Evidence an assessor asks for

- A documented continuity posture exists for every production agent.
- A failover or safe-stop has been exercised in a drill or live event within the documented cadence: at least annually and on material change.
- Loss of model is represented in the Agent Risk Register.

## Framework mappings

- NIST AI RMF: MANAGE 3
- OWASP LLM Top 10: LLM03
- SOC 2 (TSC): A1.2, A1.3
- EU AI Act: Art. 25
- Singapore: MAIGF Trusted Development and Deployment

Mappings are directional: meeting this control evidences but does not by itself satisfy the external requirement.

## Related

- Cites GOV-02 Agent Behavior Charter: https://humanintel.net/standard/gov/gov-02
- Cites GOV-03 Agent Risk Register: https://humanintel.net/standard/gov/gov-03
- Cites ACT-07 Kill switch: https://humanintel.net/standard/act/act-07
- Cites OUT-16 Fallback model assurance: https://humanintel.net/standard/out/out-16
- Cites MON-07 Agent-specific incident response: https://humanintel.net/standard/mon/mon-07
- Glossary: Continuity posture: https://humanintel.net/standard/glossary#continuity-posture

## Cite this control

Human Intelligence. HI-AAF v1.3, OUT-15 Model availability & continuity. Public review edition, July 2026. https://humanintel.net/standard/out/out-15