Is SDAIA alignment a certification?

No. It describes building against the published national strategy, ethics principles and data controls. Any formal assessment sits with the authority and with your own compliance function.

Does this force us onto specific national platforms?

Government entities integrate with national platforms where the service requires it. For everyone else the binding constraints are the data controls and residency rules, not a particular product.

Can we use models hosted outside Saudi Arabia?

It depends on the data class that reaches them. We route by class, so restricted data stays on in-country inference while open data can use whatever is best for the task.

What proves a model was tested?

The evaluation record attached to the release: the test set, the date, the scores per slice, and who signed off. Screenshots of a demo prove nothing.

National AI policy

Building AI that lines up with SDAIA's national direction

SDAIA sets the national agenda for data and AI: the strategy, the ethics principles, the data controls and the platforms government runs on. This is what that means for a system you are about to build.

Book the Data Foundation Assessment AI governance practice

The ethics principles are not decoration. They decide which decisions a model is allowed to make alone.

SDAIANDMOVision 2030

The short answer

The Saudi Data and AI Authority owns Saudi Arabia's national data and AI agenda, including the national strategy, AI ethics principles, and the data controls issued through the National Data Management Office. For a build this translates into transparency about model use, human accountability for decisions, fairness testing, documented data provenance, and hosting choices that respect national data rules.

Key takeaways

What you need to know, in a minute

Accountability sits with a person

A named human stays responsible for a consequential decision. Systems that cannot show who approved what will not pass scrutiny.

Transparency is a build requirement

Users are told when they are dealing with a model, and the reasoning behind an output can be reconstructed after the fact.

Fairness is tested, not assumed

Outcomes are measured across the groups the system touches, before deployment and again in production.

Arabic is a correctness requirement

A model that mangles Saudi names, addresses and official document formats is wrong, not merely inconvenient.

From the text to what gets built

Each line is a published expectation and what we implement against it. None of this is legal advice or a compliance verdict.

Area	What it asks for	What we build
Fairness	Avoid unjustified differences in outcome.	Slice evaluation across relevant groups, recorded before release and monitored after.
Privacy and security	Protect personal data through the AI lifecycle.	Personal data kept out of training sets, retrieved at inference under access control and logged.
Humanity and social benefit	Keep a human in the loop on consequential decisions.	Approval gates in the agent workflow, with the model's recommendation and the human decision both stored.
Reliability and safety	Behave predictably and fail safely.	Evaluation suites per release, guardrails on tool use, and a rollback path that does not need a rebuild.
Transparency and explainability	Make model involvement and reasoning visible.	Answer level citations to the source records, and a trace of which tools and documents produced an output.
Accountability	Name who answers for the system.	Model registry entries with an owner, an intended use statement and the evaluations that supported release.

Model registry and intended use

Every model in production has an owner, a stated purpose, an evaluation record and a retirement condition.

Evaluation harness

Task specific test sets in Arabic and English, run on every release, with results kept as the release evidence.

Human approval design

Deciding which actions an agent may take alone, which need approval, and where the record of that approval lives.

Arabic quality work

Saudi name handling, official document layouts, dialect in voice channels and the numerals people actually type.

Use cases

Where this applies in practice

Questions we get on SDAIA alignment

Related guides

PDPL

Most gaps we find are not legal gaps. They are missing lineage, missing classification and copies nobody owns.

Open the guide

NDMO controls

An assessment score moves when the catalogue is populated by the pipeline, not by a spreadsheet campaign.

Open the guide

NCA Essential Cybersecurity Controls

Most audit findings on a data platform are access findings. Shared accounts and forgotten service credentials, not exotic attacks.

Open the guide

SAMA expectations

The question is rarely whether the model works. It is whether you can reconstruct one decision from eighteen months ago.

Open the guide

In-country hosting

The failure is almost never the main database. It is a notebook, a support tunnel or a model API nobody classified.

Open the guide

Etimad

A bid rarely loses on the model. It loses on the appendix nobody prepared.

Open the guide

Let's put your AI ambition on the table

Sixty minutes with a senior Bilytica partner. We stress-test your plans against what actually works in your sector, agree which moves are worth funding first, and you leave with a written 90-day plan and honest numbers, whether you work with us or not.

Start the conversation