TWO REGULATORS, ONE DATA ESTATE

Two regulators, one data estate
Saudi Arabia has split jurisdiction over your data between NDMO and NCA. Your architecture has to satisfy both, on different calendars, using different methodologies.
Reading time: 9-11 minutes
There is a sentence in the Saudi National Data Management Office's published standards that I have never once heard raised in a client meeting, a vendor briefing or a conference session in this region.
It appears in section 9.15 of the National Data Management and Personal Data Protection Standards, in the Data Security and Protection domain:
"Compliance to the Data Security and Protection controls will be conducted by NCA, as per their requirements and methodology, and not as part of NDMO's annual Data Management and Personal Data Protection compliance assessment."
Read that twice, because it has consequences.

Saudi Arabia has formally divided assessment authority over the same data estate between two national regulators. NDMO, operating under the Saudi Data and Artificial Intelligence Authority, governs how data is managed, catalogued, classified, shared and protected as a national asset — across fifteen domains, pursuant to Cabinet Resolution No. 292, with entities conducting a compliance audit annually and submitting it to NDMO in the third quarter of each year. NCA governs how it is secured — through the Essential Cybersecurity Controls, the Data Cybersecurity Controls and the Cloud Cybersecurity Controls, with its own methodology and its own cadence.
Two authorities. Two methodologies. Two calendars. One data estate.
And here is the part that almost nobody has noticed, because it is buried in an appendix.
NCA did not merely accept the division. It actively dismantled part of its own control set to create it.
Appendix C of ECC-2:2024 is the change log against ECC-1:2018. Two entries in it matter more than anything else in the document.
The first records the deletion of sub-control 4-2-3-3, which had read "Entity's information hosting and storage must be inside the Kingdom of Saudi Arabia." NCA's stated rationale:
"Controls related to data localization have been transferred from the document to the National Data Management Office (NDMO) at the Saudi Data and Artificial Intelligence Authority... and entities must refer to the National Data Management Office regarding data localization before taking any action in this regard."
The second records the deletion of the old control 2-7-3, whose sub-controls had included data and information privacy. The rationale:
"For the National Data Management Office (NDMO) at the Saudi Data and Artificial Intelligence Authority mandates, entities must refer to the National Data Management Office regarding data privacy before taking any action in this regard."
So: NDMO's standards point to NCA for data security assessment. And NCA, in the change log of its own current control set, points to NDMO for both data localisation and data privacy — having removed its own controls on each.
Each authority has, correctly and deliberately, deferred to the other's competence. The organisations in the middle are the ones who have to build something that satisfies both.
⚠️ One precision point, because it matters to anyone scoping a programme. If you are looking for your in-Kingdom hosting obligation in ECC-2:2024, it is not there anymore. What remains in ECC is the operations-location rule at sub-control 4.1.3.2 — remotely-accessed managed monitoring and operations centres must be in the Kingdom. Operations location is an NCA question. Data location is an NDMO and SDAIA question. Conflating those two is the single most common scoping error I see in Saudi programmes, and after reading Appendix C you can see exactly why it happens.
The organisations in the middle are the ones who have to build something that satisfies both.
Why this breaks the way most programmes are organised
Walk into a typical Saudi public entity, or a private company inside NCA scope — as a CNI owner, operator or host, or under contractual flow-down — handling government data, and you will find two separate programmes. [Non-CNI private entities should confirm whether NCNICC-1:2025 rather than ECC is their applicable baseline.]
One sits in the data office, or with the Chief Data Officer, or in a newly created data governance function. It is working through NDMO's fifteen domains — governance, catalogue and metadata, quality, operations, document and content management, architecture and modelling, reference and master data, business intelligence, sharing and interoperability, value realisation, open data, freedom of information, classification, personal data protection, data security. It is preparing for an annual assessment. Its vocabulary is data management. Its artefacts are registers, catalogues, policies and maturity scores.
The other sits in the cybersecurity function, under the CISO. It is working through ECC-2:2024, DCC-1:2022 and CCC-2:2024. Its vocabulary is controls. Its artefacts are configurations, evidence samples and compliance tooling.
They meet at a steering committee. They produce two reports. They use two different definitions of almost every important noun.
And they are describing the same data.
The failure mode is entirely predictable, and I see it repeatedly. The data office builds a classification register. The cybersecurity function builds enforcement. Neither is wrong. But the register was built to satisfy a data management assessment, and the enforcement was built to satisfy a cybersecurity assessment, and nothing in either workstream guarantees that the labels in the register correspond to the boundaries in the network.
You end up with an entity that can pass both assessments in the same year and still cannot answer the only question that actually matters: is Secret-tier data, as classified in our register, actually subject to stronger enforcement than Confidential-tier data, in production, right now, provably?
The classification tiers are the join — and almost nobody treats them that way
There is one artefact that appears in both regimes, and it is the key to the whole problem.
NDMO's standards include a Data Classification domain. NCA's Data Cybersecurity Controls apply differentiated controls across four classification levels — Public, Confidential, Secret and Top Secret — based on sensitivity and protection need, across the full data lifecycle. The Authority is explicit that DCC compliance cannot be achieved without continuous ECC compliance.
Both regimes turn on classification. Both reference the same four-tier model derived from SDAIA's regulatory tools.
Classification is the join between the two regulators. It is the single artefact that, if built correctly once, satisfies a requirement on both sides — and if built twice, in two vocabularies, by two teams, guarantees that neither version is authoritative.
This is why I now treat classification as an architecture artefact rather than a governance artefact, and why I refuse to let it be owned exclusively by either function.
A classification scheme owned by the data office tends to be semantically rich and architecturally inert. It describes data beautifully and enforces nothing. A classification scheme owned by the security function tends to be enforceable and semantically thin — three tiers, because that is what the tool supports, mapped loosely to four regulatory levels.
What both regimes actually require is a classification model that is simultaneously the data office's register and the security function's policy input. One model. One authoritative source. Two sets of outputs.
That is an architecture decision, and it has to be made before either programme is very far along, or you are retrofitting.
The practical architecture
Here is how I sequence it when both regimes are in scope.
One. Build the classification model once, to the four regulatory tiers, as the authoritative source for both programmes.
Not three tiers. Not five. Four — Public, Confidential, Secret, Top Secret — because that is the model both regimes reference, and because a tier model that does not match the regulator's is a translation layer you will maintain forever.
Each tier gets two definitions attached to it: a semantic definition, for the data office's register and for the humans applying labels, and an enforcement definition, for the security function's policy engines. Residency permitted. Access populations permitted. Encryption and key custody required. Logging required. Retention and destruction required. Backup topology permitted.
The semantic definition answers what is this? The enforcement definition answers so what? A tier without an enforcement definition is a label, not a tier — and it will satisfy an NDMO assessor reading a register while failing an NCA assessor sampling a configuration.
Two. Build one register, with two views.
The data office needs catalogue, metadata, lineage, ownership, quality and sharing attributes. The security function needs location, replica topology, backup target, access population, enforcement point and evidence pointer.
These are views on the same record, not two records. If they become two records they diverge within a quarter, and you will spend the rest of the programme reconciling them instead of remediating anything.
Three. Derive enforcement from the tier, not from the network.
This is the point at which the architecture either works or does not. Segmentation policy, access policy, key custody and backup placement should all be derived from the classification tier of the data behind the asset — not from the network segment the asset happens to sit in, not from the application it belongs to, and not from the business unit that owns it.
When enforcement is derived from classification, two things become true that are otherwise very hard to achieve. An NCA assessor sampling any control can be shown the tier that drives it. And an NDMO assessor reviewing the classification register can be shown that the register has operational consequence rather than documentary existence.
Four. Instrument for two calendars.
NDMO's assessment is annual. NCA's methodology and cadence are its own. Your evidence generation needs to produce on demand, continuously, rather than being assembled in the weeks before whichever assessment arrives first.
This is where most programmes bleed cost. An entity preparing two separate evidence packs, manually, twice a year, against two methodologies, is spending a very large amount of money on a problem that is solved once by designing evidence into the controls and tagging each artefact with the requirement — from either regime — that it satisfies.
Five. Write down which authority owns which question and put it in front of your executive.
This sounds administrative. It is not. When a question arrives — about privacy, about transfer, about security of a particular dataset — knowing which regulator is competent, and which of your internal functions owns the response, is the difference between a two-day turnaround and a six-week internal argument.
NDMO's standards point to NCA for data security assessment. ECC-2:2024 points to NDMO for data privacy. Build that routing into your governance model explicitly, because the regulators already have.
The uncomfortable conclusion
I think the two-programme structure is the single most expensive organisational mistake being made in Saudi data governance right now, and it is being made by capable people for entirely understandable reasons. Two regulators arrived with two mandates, so two programmes were stood up. It is the obvious response.
It is also the one that guarantees you build the same classification model twice, in two vocabularies, with neither version authoritative — and then spend years reconciling them.
The regulators have already told you how they intend to divide the work. NDMO assesses data management. NCA assesses data security. Each defers to the other where the other is competent.
They split the assessment. They did not split your data.
One classification model. One register, two views. Enforcement derived from tier. Evidence instrumented once. That is the architecture that satisfies both, and it is cheaper than either programme is on its own.
Jawaid Ali is Founder and Principal Security Architect at PRAECEPTA, a specialist security and enterprise architecture practice serving regulated organisations across the Middle East and Africa. He is the author of the PRAECEPTA GCC Framework Crosswalk.
Sources: NDMO National Data Management and Personal Data Protection Standards, published at sdaia.gov.sa, issued pursuant to Cabinet Resolution No. 292; NCA Essential Cybersecurity Controls (ECC-2:2024) and Data Cybersecurity Controls (DCC-1:2022), published at nca.gov.sa. All regulatory text referenced is drawn from the issuing authority's own published documents. Verified against primary sources on 21 November 2026. Regulatory instruments change.
This article is security architecture guidance and does not constitute legal advice. Engage qualified counsel in the relevant jurisdiction for binding interpretation.


Comments