SovAIHub

AI-SOV control catalogue

Requirements you can implement and verify

Ten domains and their baseline controls connect sovereignty goals to implementation guidance, expected evidence, validation methods, and jurisdiction mappings.

Control domains

AI-SOV-01

Governance & Jurisdiction

Decision authority, ownership, governing jurisdiction, operators, and dependency accountability.

OwnershipDecision authorityJurisdictionOperatorsDependency governance

AI-SOV-02

Data Sovereignty

Control prompts, outputs, retrieval and training data across residency, access, retention, deletion, and cryptography.

Prompts and outputsRAG dataResidencyRetention and deletionEncryption

AI-SOV-03

Model Sovereignty

Know, retain, license, host, replace, and govern model artifacts through their lifecycle.

OriginWeightsLicensingProvenanceReplacement

AI-SOV-04

Inference Sovereignty

Keep inference execution, endpoints, isolation, and control-plane dependencies inside the approved boundary.

Inference locationExternal APIsControl planesIsolationRuntime dependencies

AI-SOV-05

Infrastructure Sovereignty

Control compute, accelerators, clusters, storage, networks, and administrative dependencies.

Compute and GPUKubernetes/OpenShiftStorageNetworkingAdministration

AI-SOV-06

Artifact & Supply Chain Sovereignty

Import, verify, approve, store, and reproduce every critical model and software dependency.

RegistriesChecksums and signaturesProvenance and SBOMControlled importsApproved artifacts

AI-SOV-07

Agent & Tool Sovereignty

Bound agent permissions, tools, external services, approvals, sandboxes, and outbound actions.

Tool catalogueMCP integrationsPermissionsHuman approvalOutbound calls

AI-SOV-08

Operational Sovereignty

Operate, support, upgrade, recover, transfer knowledge, and substitute components independently.

Support and upgradesRecoveryIncidentsVendor lossSubstitution

AI-SOV-09

Security & Assurance

Enforce identity, secrets, key management, monitoring, testing, auditability, and configuration assurance.

IAMSecrets and keysLoggingVulnerability managementTesting

AI-SOV-10

Network & Air-Gap Sovereignty

Control Internet, telemetry, DNS, updates, endpoints, and disconnected operations.

Internet dependencyEgressTelemetryDisconnected operationTransfer zones

Baseline catalogue

Assurance controls

These controls are deliberately technology-neutral. Product mappings show contributions; they never pre-award a result.

AI-GOV-001 · Governance & Jurisdiction

Owned sovereignty boundary

Level 1

The system must have an approved sovereign boundary, named decision authority, operators, and dependency owners.

A location claim is insufficient unless authority and external dependencies are explicit.

Implementation guidance

Maintain a versioned system definition and responsibility map with approval and reassessment triggers.

Expected evidence

  • Approved system definition
  • Responsibility matrix
  • Dependency register

Validation methods

documentaryindependent review
EU SOV-1EU SOV-2

AI-DAT-001 · Data Sovereignty

Controlled AI data lifecycle

Level 2

Prompts, outputs, retrieval data, and training data must remain within approved locations, identities, retention, and deletion rules.

Data control covers movement and lifecycle, not residency alone.

Implementation guidance

Propagate classification and permissions through ingestion, retrieval, logs, caches, backups, and deletion.

Expected evidence

  • Data-flow map
  • Storage configuration
  • Permission and deletion tests
  • Key ownership record

Validation methods

configurationtechnical testdocumentary
EU SOV-2EU SOV-3EU SOV-7

AI-MOD-001 · Model Sovereignty

Independent model artifact custody

Level 2

The organization must retain authorized access to identified model artifacts and a tested replacement path appropriate to the target level.

API access or a mutable model name is not independent custody.

Implementation guidance

Record model identity, license, provenance, evaluation, package, internal location, approval, and retirement path.

Expected evidence

  • Model manifest and hash
  • License decision
  • Internal repository record
  • Replacement test

Validation methods

configurationdocumentarytechnical test
EU SOV-3EU SOV-5EU SOV-6

AI-INF-001 · Inference Sovereignty

Sovereign inference boundary

Level 2

AI inference must execute within the defined sovereign execution boundary without a mandatory external inference API at Level 2 or above.

Runtime endpoints, routing, and control planes are part of the inference boundary.

Implementation guidance

Use customer-controlled endpoints, deny unapproved routes, and test inference during external-service loss.

Expected evidence

  • Deployment manifest
  • Inference configuration
  • Network policy
  • Egress and dependency-loss test

Validation methods

configurationtechnical test
EU SOV-3EU SOV-4EU SOV-6

AI-PLT-001 · Infrastructure Sovereignty

Recoverable platform boundary

Level 3

Critical compute, storage, network, scheduling, and administration must be operable and recoverable under the target dependency conditions.

Owning hardware does not prove recovery or administrative independence.

Implementation guidance

Define the tested platform bill of materials, privileged paths, capacity envelope, backup, restore, and replacement procedure.

Expected evidence

  • Platform configuration
  • Administrative access map
  • Capacity test
  • Recovery exercise

Validation methods

configurationtechnical testoperational exercise
EU SOV-4EU SOV-6EU SOV-7EU SOV-8

AI-SUP-001 · Artifact & Supply Chain Sovereignty

Controlled artifact promotion

Level 3

Critical software and model artifacts must enter through a verified, approved, traceable promotion path and remain recoverable from authoritative sources.

A disconnected runtime still depends on its import, update, and recovery supply chain.

Implementation guidance

Separate acquisition, quarantine, approval, internal repositories, offline builds, revocation, and rollback.

Expected evidence

  • Artifact manifest
  • Checksums or signatures
  • SBOM where applicable
  • Approval log
  • Offline rebuild test

Validation methods

configurationdocumentarytechnical test
EU SOV-4EU SOV-5EU SOV-6EU SOV-7

AI-AGT-001 · Agent & Tool Sovereignty

Bounded agent and tool authority

Level 4

Agents may use only approved tools, identities, destinations, permissions, and human approval paths within the declared boundary.

An external tool can reintroduce data, jurisdiction, and continuity dependencies.

Implementation guidance

Maintain a versioned tool catalogue and enforce deny-by-default authorization, sandboxing, egress, approval, and action receipts.

Expected evidence

  • Tool registry
  • Authorization policy
  • Sandbox and egress tests
  • Approval and action receipts

Validation methods

configurationtechnical testindependent review
EU SOV-3EU SOV-4EU SOV-7

AI-OPS-001 · Operational Sovereignty

Independent operation and recovery

Level 3

The organization must be able to operate, diagnose, update, recover, and evolve the AI capability without continuous original-vendor involvement.

Operational sovereignty is demonstrated through exercised capability and retained knowledge.

Implementation guidance

Test vendor-loss, operator handover, rollback, restore, component substitution, and controlled update scenarios.

Expected evidence

  • Operating runbook
  • Recovery record
  • Handover record
  • Vendor-loss and substitution exercise

Validation methods

operational exercisedocumentaryindependent review
EU SOV-1EU SOV-4EU SOV-6

AI-SEC-001 · Security & Assurance

Local security and assurance authority

Level 1

Identity, secrets, keys, logging, testing, and security decisions must have controlled authorities and reviewable evidence.

Security supports sovereignty but does not replace jurisdictional or operational independence.

Implementation guidance

Bind every privileged action and release to identity, policy, configuration, test, and retained evidence.

Expected evidence

  • IAM and privilege map
  • Key and secret ownership
  • Security test results
  • Audit configuration

Validation methods

configurationtechnical testindependent review
EU SOV-2EU SOV-7

AI-NET-001 · Network & Air-Gap Sovereignty

Controlled runtime connectivity

Level 2

Runtime Internet, telemetry, DNS, licensing, update, and external endpoint dependencies must be declared, constrained, and compatible with the target level.

Level 4 requires no mandatory runtime Internet, outbound telemetry, or online license validation.

Implementation guidance

Inventory destinations and failure behavior, enforce egress policy, and exercise disconnected operation.

Expected evidence

  • Network dependency inventory
  • Firewall and DNS policy
  • Telemetry configuration
  • Disconnected-runtime test

Validation methods

configurationtechnical testoperational exercise
EU SOV-3EU SOV-4EU SOV-7