Scale AI and Oracle publish a blueprint for keeping government AI workloads in-country

Government teams have a new deployment blueprint designed to keep their AI data and model processing in-country. Scale AI and Oracle have published a joint reference architecture for Scale GenAI Portfolio on OCI Dedicated Region. Scale says it combines agent-building tools with a defined regional boundary and customers’ existing security controls.

The publication sets out how Scale’s software runs on Oracle Cloud Infrastructure’s Dedicated Region offering. It covers connecting data, building and running agents, and evaluating their performance. Scale presents it as a reusable deployment pattern, intended to spare teams from assembling those pieces from scratch.

A boundary around prompts, documents and answers

Scale describes the architecture as a sovereign AI setup: data, models and applications stay in the country. Its model-serving arrangement is central to that claim. Models run in-region, and prompts, documents retrieved to help answer a request, and generated outputs remain inside the customer’s OCI Dedicated Region environment.

The security design also relies on systems the customer already uses. Scale says GenAI Portfolio uses the tenant’s existing identity, access-control and auditing systems, rather than requiring another security layer to set up and manage. The reference therefore addresses both where information is processed and which controls govern access to it.

Qatar experience, followed by a staged rollout

Scale says the architecture reflects its work building and operating AI applications for government entities in Qatar, including deployments on OCI Dedicated Region in Doha. That existing work is the basis for its description of the blueprint as production-ready; the publication is a reference architecture, not an announcement of a new government contract.

The offering extends beyond initial agent development. Scale describes tools for monitoring agents and improving them over time, alongside domain expertise, language support including Arabic, and delivery teams that remain involved after deployment. Its suggested adoption process starts with one application, rather than a broad rollout:

1. Choose a use case with a clear owner and a measurable outcome. That gives the deployment a specific task and someone responsible for it.
2. Map its data sources, access requirements, evaluation criteria and deployment needs to the reference architecture on OCI Dedicated Region.
3. Define acceptable performance before launch, together with how the team will monitor the application afterward. Evaluation is part of the deployment plan, not just a later activity.
4. Deploy inside the customer’s OCI Dedicated Region environment, then reuse the infrastructure and controls for the next application.