AnswersPublic sector

How should government agencies deploy sovereign AI?

In short

Government agencies deploy sovereign AI by keeping control of four things: where the data and models run, which models they use, what every agent is allowed to do, and the ability to change suppliers without losing what they built. In practice that means deployment in the agency's own environment, a choice of models, approvals and audit records on every action, and contract terms that keep the service replaceable.

What sovereignty requires

RequirementWhat it means in practice
ResidencyData, models and logs stay in the agency's chosen environment and jurisdiction.
Model choiceMore than one model family, including open-weight models the agency can run itself.
Control of actionsAgents act only within approved permissions, with a person approving consequential steps.
AuditEvery action recorded and exportable for oversight and records requests.
ReplaceabilityInstructions, evaluations and trained models stay with the agency if a supplier changes.

Where agencies start

Citizen-facing service is the most visible start: answering questions, guiding applications and routing cases, in plain language and in the languages residents speak. Internal casework, permitting, procurement and records requests follow, each measured against current processing times.

Procure for change

Model suppliers change terms, retire models and change availability. Agencies that write portability, export and model choice into the contract keep their service running when that happens. Fig runs CosmOS inside the agency's own environment or Fig's, with open-weight and frontier models, approvals on consequential actions and an exportable audit trail.

Frequently asked questions

What is sovereign AI in government?

AI an agency controls: where it runs, which models it uses, what it may do and the ability to change suppliers without losing its work.

Can government AI run on open-weight models?

Yes. Open-weight models can run inside the agency's own environment, alongside frontier models where policy allows.

How should agencies handle AI audit and records?

Record every agent action with its inputs, outputs and approvals, and keep those records exportable for oversight and records requests.

See it running on your company's real work.

More answers →

Be the next big thing

Dream big, build fast, and grow far with Fig

Fig Desktop for Mac and Windows