Government — Sovereign AI for public institutions

002 / The Principle

A state cannot govern AI it does not operate.

Government AI is not a cost-reduction story. It is a sovereignty story. Every foreign-hosted government AI workload is a workflow whose governance has been exported. The data, the model, the audit record — all subject to a legal regime the institution cannot write, cannot read in its own language, and cannot enforce on its own terms.

This is not a theoretical argument about jurisdiction. It is the practical question of whether public institutions in Latin America can build AI capability without creating permanent dependence on foreign-controlled infrastructure.

01 / JURISDICTION

The law that governs is the law you know.

Workloads operating under foreign CLOUD Act exposure create permanent legal risk. Sovereign deployments close that exposure by architecture.

02 / CAPABILITY

Institutional capability cannot be rented.

A country that does not operate its own AI stack does not have AI capability. It has AI access — a temporary state that can be revoked by policy, pricing, or procurement.

03 / ACCOUNTABILITY

The citizen's recourse ends at the border.

When a public institution's AI makes an error affecting a citizen, the accountability path must terminate inside the country. Foreign-hosted workflows break that path.

003 / Use Cases

Where sovereign AI actually runs.

Public-sector AI workloads live on a spectrum — from the highly sensitive (security, social program administration) to the more routine (institutional knowledge, citizen communications). Saptiva AI handles the full range under a single platform and a single compliance framework.

01

National-scale language models.

Sovereign LLMs trained and hosted inside the country, validated by infrastructure partners, available to public institutions and regulated enterprises through the Saptiva AI platform. KAL is the first. It is the template for the next ones across the region.

NATIONAL SCALEIN-COUNTRYVALIDATED

02

Citizen service automation.

First-tier citizen inquiries — program eligibility, procedural status, document requirements — resolved automatically with clear escalation paths to human staff. Public-facing services improve in responsiveness without leaving the country's digital borders.

MULTI-CHANNELHUMAN ESCALATIONFULL AUDIT

03

Document processing at institutional scale.

High-volume government document flows — permits, certifications, applications, archives — processed with the same audit and residency posture as the rest of the platform. No document leaves the jurisdiction. Exception cases route to human reviewers with structured reasoning attached.

RESIDENTSTRUCTURED OUTPUTEXCEPTION ROUTING

04

Institutional knowledge copilots.

Private AI systems trained on the institution's policies, procedures, and historical decisions. Role-based access, in-country hosting, full retrieval audit. The institution's knowledge stays the institution's knowledge.

PRIVATE RAGROLE-BASEDIN-COUNTRY

05

Regulatory oversight support.

AI tooling for regulatory and supervisory agencies — the institutions whose job is to oversee the banks, insurers, and markets we serve. The same platform powering compliance inside regulated firms can equip the regulators inspecting them.

SUPERVISORYEXPLAINABLEANALYST ASSIST

06

Critical infrastructure and defense-adjacent workloads.

Air-gapped or fully isolated deployments for the institutions whose connectivity requirement is itself a risk. Saptiva AI's air-gapped mode is the same platform running under zero external connectivity at runtime.

AIR-GAPPEDISOLATEDFULL AUDIT

004 / Sovereignty In Practice

Sovereignty is a deployment decision, not a marketing claim.

Any vendor can put the word "sovereign" on a landing page. What matters is what the deployment actually looks like in practice — who hosts the compute, who holds the keys, who writes the policy, who retains the audit. Saptiva AI's sovereign deployments pass these four tests:

01

The compute lives inside the country.

Not a "region" named after the country while the data plane lives elsewhere. Physical infrastructure inside the jurisdictional border, operated under local law.

02

The encryption keys are held by the institution.

The institution controls the keys. The vendor cannot decrypt data, even under compulsion from a foreign legal process. This is the technical closure that makes sovereignty real.

03

The policy is written by the institution.

FrIdA evaluates the institution's policy file, not a vendor-defined approximation of it. If the framework changes, the policy file changes. The platform does not.

04

The audit record stays under institutional custody.

Signed, immutable, in-country. The institution — not the vendor — is the authoritative source of the record of what its AI did, on what data, under whose authorization.

005 / Customer

Anchored by . Mexico's first national-scale LLM.

GOVERNMENT · MEXICO · IN PRODUCTION

Mexico's first national-scale LLM. Built in collaboration with the Mexican government. Validated by NVIDIA. Operating under Mexican jurisdiction.

KAL is the public evidence that sovereign AI is not a theoretical capability in Latin America — it is a deployed one. A three-party structure between Saptiva AI, the Mexican government, and NVIDIA produced a model and a platform layer that Mexican institutions can use without routing their governance through a foreign jurisdiction.

For peer institutions across the region asking the same question — under whose law does our AI run? — KAL is the working reference.

Read the deployment →

Sovereign AI. Under your law.