# PDPL, NCA and SDAIA: what they ask of an AI project

Compliance starts with the design. PDPL, NCA's Essential Cybersecurity Controls and SDAIA's AI Ethics Principles shape almost every AI project in the Kingdom. Here is what each asks of you in practice.

Octopus research team · 16 November 2026 · Governance & regulatory readiness, Data, platforms & sourcing

When an AI project stalls, the compliance review is often where it happened. The review came after the design was fixed, found a problem, and sent the team back to the start. Three frameworks shape almost every AI project in the Kingdom. Each turns into a short list of decisions that belong in the first workshop, not after go-live.

This is a practical guide to use from the first workshop; your legal and security teams sign off.

## The Personal Data Protection Law (PDPL)

The PDPL came into force on 14 September 2023 [1], and SDAIA gave organisations a year to bring their processing into line [2]. AI projects touch more personal data than teams expect: names and ID numbers in scanned documents, employee records, customer messages.

In practice the law shapes three decisions.

1. Purpose. Process only what the use case needs, and be able to explain why each field is there.
2. Transfers. Personal data can leave the Kingdom only under specific conditions and safeguards, with a risk assessment [3] [4]. Calling a model hosted abroad can count as a transfer.
3. Breaches. The implementing regulation requires notifying SDAIA within 72 hours of becoming aware of a breach that could cause harm [3] [5]. Your system's logs and data flows have to support that.

### Five rights the system has to serve

The law gives individuals five rights: to be told the legal basis and purpose of collection, to access their personal data, to obtain a copy in a readable and clear format, to have it corrected, completed or updated, and to request its destruction [11]. The Implementing Regulation sets the clock: a controller answers a request within 30 days, and can extend that by up to 30 more days if it tells the person in advance [3].

For an AI system, the hard part is finding the data. Personal data can sit in prompts, conversation logs, search indexes built for retrieval, and any dataset used to tune a model. Decide at design time how a request to access or destroy data reaches each of those places, and who carries it out.

### When an impact assessment is required

The Implementing Regulation requires an impact assessment before certain kinds of processing [3]. The triggers include processing sensitive data; collecting, comparing or linking two or more datasets from different sources; large-scale or repeated processing that relies on newly adopted technologies or makes decisions based on automated processing; and any product or service likely to cause serious harm to people's privacy.

Many AI use cases meet at least one of these triggers. A model that links customer records with call transcripts links datasets. A system that ranks applications makes decisions from automated processing. Run the assessment during design, while its findings can still change the design.

### Records, a data protection officer and registration

Controllers keep a record of their processing activities for as long as the processing runs, and for five years after it ends [3]. Each new AI use case that touches personal data adds an entry to that record.

A personal data protection officer is required where the controller is a public entity providing services that involve processing personal data on a large scale, where its primary activities require regular and systematic monitoring of individuals, or where its core activities rest on processing sensitive data [3]. Public entities, controllers whose main activity is processing personal data, and controllers that process sensitive data must also register on SDAIA's National Data Governance Platform [12].

### Transfers, in more detail

The transfer regulation, in its second version of August 2024, sets out how personal data can leave the Kingdom [13]. SDAIA publishes a list of countries and international organisations it considers to offer an appropriate level of protection, and reviews it every four years or as needed. Where a destination is not on that list, the transfer relies on safeguards: standard contractual clauses, binding common rules or a certificate of accreditation [13] [4].

A risk assessment is required before a transfer that relies on those safeguards, and before transferring or disclosing sensitive data on a continuous or widespread basis [13]. If your model provider processes prompts outside the Kingdom, this is the assessment your legal team will ask for, so collect the provider's data flows early.

### What non-compliance costs

Violations can draw a warning or a fine of up to SAR 5 million, which can be doubled for a repeat violation [11]. Disclosing or publishing sensitive data in breach of the law, with intent to harm the person or for personal benefit, carries up to two years in prison, a fine of up to SAR 3 million, or both [11].

## NCA's Essential Cybersecurity Controls (ECC)

The ECC apply to government entities and the companies affiliated with them, and to private organisations that own, operate or host critical national infrastructure. NCA encourages every other organisation to use them as good practice. The current version, ECC-2:2024, has four main domains: governance, defense, resilience, and third-party and cloud computing [6] [7].

In numbers, ECC-2:2024 has 4 main domains, 28 subdomains, 108 main controls and 92 subcontrols [6]. The Cloud Cybersecurity Controls, CCC-2:2024, have 4 main domains and 24 subdomains, with 37 main controls for cloud service providers and 18 for cloud service tenants, the organisations that use the cloud [14]. Where the cloud controls apply to you, an AI system hosted on a cloud platform usually puts you on the tenant list.

For an AI project, the third-party and cloud domain usually matters most. A model provider, a hosting platform and an integration partner are all third parties. If you use cloud hosting, the cloud controls apply. They include protecting your data according to its classification, and getting it back in a usable form when the service ends.

Decide early whether the system runs on-premises or with a cloud provider in the Kingdom. That choice changes what you document, which suppliers you assess, and which contract clauses you need.

## SDAIA's AI Ethics Principles

The principles apply to anyone designing, developing, deploying or using AI systems in the Kingdom [8]. There are seven.

- Fairness
- Privacy and security
- Humanity
- Social and environmental benefits
- Reliability and safety
- Transparency and explainability
- Accountability and responsibility

The framework also sorts AI systems by risk. Systems with little or no risk, such as spam filters, carry no restrictions. Limited-risk systems are subject to the principles. High-risk systems that affect basic rights need conformity assessments before and after deployment. Systems with unacceptable risk are not allowed [8].

Classify each use case early, because the class sets how much testing, documentation and human oversight it needs. SDAIA also offers a self-assessment tool for AI ethics on the National Data Governance Platform [9].

The principles also expect named roles. The edition now on SDAIA's website asks entities to designate the people responsible for AI ethics: the head of the entity or the chief data officer, a chief compliance officer as the strategic lead, and a Responsible AI Officer as the operational lead [15]. If your organisation has none of these yet, name them before the first high-risk use case goes live.

## What is coming next

SDAIA put a draft Responsible AI Policy out for public consultation in April and May 2026. According to a November 2026 legal briefing, the draft would make AI governance binding: risk tiers, registration of AI systems, safety reports before high-risk deployment, and annual third-party audits for high-risk systems [10]. It is still a draft, and the final text may change. The steps it points to are worth taking now either way: an inventory of your AI systems, a risk class for each, and human oversight where it matters.

SDAIA also consulted on amendments to the PDPL Implementing Regulation from 6 October to 5 November 2026. According to a legal briefing, the drafts would require an impact assessment where emerging technologies are used or processing leads to automated decisions, and would require breach notice within 72 hours without the current test of likely harm [16]. They are drafts, not rules in force.

> Rules are coming. Build for them now, not after.
> (Octopus)

**Six items for your first project workshop**

- A data map: which personal data the use case touches, and why each field is needed.
- A hosting decision: on-premises, a cloud in the Kingdom, or a justified transfer with its risk assessment.
- Your ECC position: whether your organisation is in scope, and which third-party and cloud controls apply.
- A risk class for the use case under SDAIA's framework, and the human oversight it requires.
- An impact assessment where the use case meets a trigger, and its entry in your record of processing.
- A breach path: who detects it, who decides, and how SDAIA is notified within 72 hours when required.

Across all three frameworks, the questions are cheap to answer in the first workshop and expensive to answer after go-live. Treat them as design inputs, and compliance stops being the step that sends a project back to the start.

This is the work of our practice in AI governance, risk and regulatory readiness. Hosting and data decisions sit with our practice in data, sovereign platforms and technology sourcing.

## Sources

Figures are quoted as published. Forecasts and drafts are labelled.

1. [Saudi Press Agency, SDAIA announces the PDPL in force, Sep 2023](https://www.spa.gov.sa/14233f0780e)
2. [Saudi Press Agency, one-year period to comply with the PDPL](https://www.spa.gov.sa/N2136222)
3. [SDAIA, PDPL Implementing Regulation](https://sdaia.gov.sa/en/SDAIA/about/Documents/ImplementingRegulation.pdf)
4. [SDAIA, Standard Contractual Clauses for personal data transfer](https://sdaia.gov.sa/Documents/StandardContractualClausesForPersonalDataTransferEN.pdf)
5. [SDAIA, personal data breach notification service](https://dgp.sdaia.gov.sa/wps/portal/pdp/services/personaldatabreachnotification)
6. [NCA, Essential Cybersecurity Controls ECC-2:2024 (English)](https://cdn.nca.gov.sa/api/files/public/upload/86e09090-44e4-481f-bc28-355673607654_ECC--2024-EN.pdf)
7. [NCA, Essential Cybersecurity Controls ECC-2:2024 (Arabic)](https://cdn.nca.gov.sa/api/files/public/upload/29a9e86a-595f-4af9-8db5-88715a458a14_ECC-2-2024---NCA.pdf)
8. [SDAIA, AI Ethics Principles, Sep 2023](https://dgp.sdaia.gov.sa/wps/wcm/connect/4c56ed1c-1b82-447d-ac29-638f5f99c12e/ai-principles-EN.pdf)
9. [Saudi Press Agency, SDAIA AI ethics self-assessment tool](https://www.spa.gov.sa/N2210858)
10. [AS&H Clifford Chance, briefing on SDAIA's draft Responsible AI Policy, 4 Nov 2026 (draft; public consultation April to May 2026)](https://www.cliffordchance.com/content/dam/cliffordchance/briefings/2026/11/responsible-ai-policy-shift-guiding-regime-supervisory-one.pdf)
11. [SDAIA, Personal Data Protection Law (English text, as amended): rights Art. 4, penalties Arts. 35 and 36](https://sdaia.gov.sa/en/SDAIA/about/Documents/PersonalDataProtectionLaw.pdf)
12. [SDAIA, Rules Governing the National Register of Controllers within the Kingdom, Art. 2](https://sdaia.gov.sa/Documents/TheRulesGoverningTheNationalRegisterOfControllersWithinTheKingdomPublicEN.pdf)
13. [SDAIA, Regulation on Personal Data Transfer outside the Kingdom, version 2.0, August 2024: adequacy Art. 3, safeguards Art. 4, risk assessment Art. 7](https://sdaia.gov.sa/Documents/RegulationonPersonalDataEN.pdf)
14. [NCA, Cloud Cybersecurity Controls CCC-2:2024](https://cdn.nca.gov.sa/api/files/public/upload/6d5408a3-d8e6-4e96-963b-2c7198e5b7c2_CCC-2-2024-EN-.pdf)
15. [SDAIA, AI Ethics Principles, edition dated May 2025 (SDAIA-P114E): AI ethics roles](https://sdaia.gov.sa/en/SDAIA/about/Documents/ai-principles.pdf)
16. [Clyde & Co, SDAIA publishes draft amendments to the Implementing Regulations of the PDPL, 8 October 2026 (legal briefing; draft amendments)](https://www.clydeco.com/en/insights/2026/10/sdaia-publishes-draft-amendments-to-the-pdpl)

## Related practices

- [AI governance, risk and regulatory readiness](https://pioctopus.ai/en/services/governance-risk-compliance)
- [Data, sovereign platforms and technology sourcing](https://pioctopus.ai/en/services/data-sovereign-platforms)

## Put it to work.

The tool that goes with what you just read.

[SDAIA Compass Worksheet](https://pioctopus.ai/en/toolkit/sdaia-compass-worksheet): Map your AI initiatives to the seven components of SDAIA's AI Adoption Framework, and track pilot-to-production.

## Keep reading

- [AI governance in Saudi Arabia: a practical guide](https://pioctopus.ai/en/insights/ai-governance-saudi-arabia): Which rules apply to an AI system in the Kingdom, who publishes each one, what is in force and what is still a draft, and where to start.
- [How a Saudi government entity starts with AI](https://pioctopus.ai/en/insights/ai-adoption-government-entities): Start from SDAIA's national references, set up an AI office with a named person accountable, measure readiness in four weeks and choose the first three use cases.
- [SDAIA's National AI Index: what it measures and how to prepare](https://pioctopus.ai/en/insights/national-ai-index-saudi-arabia): The National AI Index scores government entities on three pillars, seven sub-pillars and 23 domains, with evidence behind every answer. The structure, the six maturity levels, the measurement cycle and the evidence to keep.

## Want to talk about your own operations?

- [Book a session](https://pioctopus.ai/en/contact)
- [Get the worksheet](https://pioctopus.ai/en/toolkit/sdaia-compass-worksheet)

---
Web page: https://pioctopus.ai/en/insights/pdpl-nca-sdaia-in-practice
Arabic version: https://pioctopus.ai/insights/pdpl-nca-sdaia-in-practice.md
Index for AI agents: https://pioctopus.ai/llms.txt
