SovAIHub Assurance Levels
SovAIHub Levels 0–4 are an independent, AI-native maturity model. They are not European Commission SEAL levels, a certification, or an awarded status.
A level applies only when all required controls are implemented and supported by sufficient evidence. Partial coverage should be reported by domain, not rounded up.
The AI capability depends materially on externally controlled inference, infrastructure, artifacts, or operations.
Dependencies must first be identified and governed
Providers, access, logging, data rules, residency, and operating responsibilities are explicitly governed.
2 baseline controls required
Sensitive data and inference remain inside the approved boundary without a mandatory external inference API.
6 baseline controls required
The organization can operate, update, recover, and substitute critical components without continuous original-vendor involvement.
9 baseline controls required
The complete AI capability can be securely operated, recovered, and evolved without mandatory runtime Internet or vendor dependency.
10 baseline controls required
Hard caps
A strong score elsewhere cannot compensate for a dependency that defeats the claimed operating state.
The workload cannot perform required inference without an externally controlled API.
Inference is outside the sovereign execution boundary.
A third-party SaaS control plane is required to start, route, authorize, or recover runtime service.
Independent operation is not possible.
Runtime operation requires outbound telemetry that cannot be disabled or locally terminated.
Disconnected operation cannot be demonstrated.
Critical runtime components stop without recurring external license validation.
Level 4 continuity depends on network and vendor availability.
The organization cannot retain or restore the critical model artifact under its authorization.
Model continuity and substitution are constrained.
Critical artifacts, configuration, data, keys, or operating knowledge cannot be restored.
Operational sovereignty requires tested recovery.
Required AI service cannot continue while public Internet access is denied.
Air-gap-capable Level 4 operation is impossible.
A material agent outcome depends on an externally controlled tool without an approved local substitute or fail-safe path.
The agent capability cannot be operated autonomously.