Sovereign-routed governance for a multi-model world.
Most institutions will not depend on a single national AI ecosystem. SAG-OS is the neutral operating layer that governs approved models across jurisdictions, controls what may cross a border, and produces an evidence trail regulators can defend.
You choose the AI. Cabier orchestrates the trust.
Module 3 · Cross-Border AI Gateway
Resolve the four gateway questions before the packet moves.
This simulator applies a simplified rule set derived from the SAG-OS policy engine. Full test procedures, jurisdiction-specific deltas, and legal-basis templates are released under engagement.
Select an origin, host and data class, then click Resolve gateway to see a signed decision reference.
Gate 1
Can this prompt legally leave the country of origin?
Gate 2
Can inference be performed offshore for this data class?
Gate 3
Can these model weights be imported into this jurisdiction?
Gate 4
Can embeddings cross this border, and under which legal basis?
Module 1 · AI Model Registry
Every approved model. Every version. Every deployment.
A single registry of AI systems approved for a business function — vendor, family, version, fine-tune, evaluation state, jurisdictional approval, accountable owner, effective and end dates. Nothing runs in production without a registry entry.
Module 2 · Sovereign AI Policy Engine
Per-jurisdiction rules, one control plane.
Country-specific policy packs mapped to the same kernel — no forked codebase, no shadow rulebooks. China is treated as an observed jurisdiction for cross-border cases; SAG-OS does not deploy in-country in the PRC.
United States
SR 11-7 · NIST AI RMF · State AI laws · Executive orders
Canada
AIDA (proposed) · OSFI E-23 · PIPEDA · AI-for-All strategy
United Kingdom
FCA · PRA · ICO · AI regulation white paper
European Union
EU AI Act (Art. 5/6/50/53/55) · GDPR · DORA
UAE
UAE AI strategy · CBUAE guidance · DIFC / ADGM
Saudi Arabia
SDAIA AI Ethics · PDPL · SAMA guidance
Singapore
MAS FEAT · Model AI Governance Framework · PDPA
Japan
JFSA · METI AI governance guidelines · APPI
Hong Kong
HKMA · SFC · PCPD
Australia
APRA · OAIC · voluntary AI safety standard
India
DPDP Act · MeitY AI advisories · RBI
South Korea
AI Basic Act · PIPA · FSC
Africa (selected)
SARB · CBN · KDPC · continental AI strategy
Latin America (selected)
Brazil ANPD · Mexico INAI · sectoral rules
Observed: China
PIPL · CAC generative-AI measures · sovereign export controls
Module 4 · AI Evidence Vault
Prompt, output, decision — one signed record.
The AI Evidence Vault is the public-facing view of the Assurance Kernel's evidence spine, applied specifically to AI systems. Six item classes, one signed audit ID per event.
Every record is tied to an accountable executive, a jurisdictional envelope, a model registry entry and a Trust Gate outcome — reproducible on demand.
Vault explorer — published sample
Search an illustrative set of evidence records: the artefact held, the control it satisfies, the reviewer who asserted it and the system it is bound to.
Credit-memo faithfulness eval, 480-file holdout set
Credit adjudication assistant · Evaluation results · reviewed by Independent validation
Adjudicator amendment log with sign-off scope
Credit adjudication assistant · Human oversight records · reviewed by Chief Credit Officer
Alert-rate dispersion analysis by segment
Transaction monitoring triage · Fairness and bias testing · reviewed by Model risk
Prompt-injection and tool-abuse test report
Contract review agent · Red-team findings · reviewed by Offensive security
Model provider training-data and eval declaration
Contract review agent · Vendor attestations · reviewed by Third-party risk
Citation-fidelity measurement on KYC pack summaries
Client onboarding summariser · Grounding and factuality · reviewed by Independent validation
Precision/recall regression against prior release
Market surveillance anomaly model · Evaluation results · reviewed by Model risk
Compliance review queue with reviewer assertions
Supervisory correspondence drafter · Human oversight records · reviewed by Head of Regulatory Affairs
Sovereign AI Assurance
Six commitments a sovereign deployment has to meet.
SAG-OS is the policy engine. Sovereign AI Assurance is what the engine is held to — the commitments a group, a regulator or a sovereign operator can test us against.
Dual-regulator satisfaction
One control, evidenced once, accepted by home and host supervisors — with the delta each regime adds recorded explicitly rather than negotiated case by case.
Sovereign deployment options
Multi-tenant, dedicated VPC or on-premise sovereign enclave. No mandatory US data storage; residency is a deployment parameter, not a caveat.
Cross-border evidence without data transfer
Assurance artefacts and attestations cross the border; the underlying records do not. Each crossing carries a signed legal basis.
Jurisdiction-specific AI control libraries
Per-jurisdiction control sets with test procedures, evidence artefacts and deltas — published as a sample, held in full within engagement.
AI supply-chain and model provenance
Provenance, licensing and third-party posture recorded for every model, dataset and inference provider in the estate, including those inside vendor stacks.
Continuous multi-model assurance
Drift, quality, incident and policy-violation telemetry across every model in the estate, continuously — not an annual attestation.
Guardrails