Data discovery and RoPA
Finding where personal data actually lives — including the exports, shared drives and shadow databases nobody listed — and building records of processing that reflect reality rather than intention.
Discovery · RoPA · Data mapMost organisations we assess have the documents. What they cannot do is list every system holding personal data, execute a deletion request across all of them, or prove a retention rule was ever enforced. That gap is where enforcement risk actually lives.
PDPL governs how organisations handle personal data relating to individuals in the Kingdom, and it reaches organisations outside Saudi Arabia that process such data. SDAIA supervises it. The obligations are recognisable to anyone who has worked with modern privacy regimes: have a lawful reason to process, tell people what you are doing, honour their rights, keep the data secure, do not keep it forever, and report breaches.
None of that is controversial, and none of it is where organisations fail.
They fail on execution. A deletion request arrives and touches nine systems, three of which nobody remembered. A retention policy says eighteen months and no job enforces it. Consent is captured on the website and ignored by the call centre. The privacy notice is accurate about a process that changed two years ago.
This is engineering work, not drafting work — which is precisely why it tends to sit unowned between legal and IT.
These come up in nearly every assessment. Several are repeated confidently by vendors, which is part of why they persist.
| Commonly said | More accurate |
|---|---|
| "All personal data must stay inside Saudi Arabia." | The position shifted with the 2023 amendments and subsequent regulations. Cross-border transfer is permitted subject to conditions, which can include assessing the transfer. Sector regulators may still impose stricter rules, so the answer depends on your sector and the specific transfer. |
| "We have consent, so we can process anything." | Consent is one lawful basis among several, and it is tied to the purpose it was given for. Reusing data collected for one purpose to do something else is a purpose-limitation problem regardless of the consent held. |
| "We published a privacy notice, so we are compliant." | A notice that does not match what the systems actually do is worse than none: it is documented evidence of a discrepancy. Notices need to be verified against reality periodically. |
| "Encryption means we are secure and therefore compliant." | Security is one obligation. Lawful basis, minimisation, retention, rights and transparency are separate, and an encrypted database kept for nine years past its purpose is still a retention failure. |
| "We have 72 hours after a breach, which is plenty." | Seventy-two hours runs from awareness. Organisations that take three weeks to detect an incident have already missed it. The real requirement behind the rule is detection capability. |
| "Anonymised data takes us outside the law." | True anonymisation is a high bar and harder than most teams assume. Pseudonymised data that can be re-identified with information you still hold remains personal data. |
Most privacy programmes in the Kingdom were designed before their organisation deployed AI. They now govern a set of activities that raise questions the documents do not answer, and the gap is widening faster than the paperwork is being revised.
Training data is processing. Using customer records to train a model is a processing activity requiring a lawful basis and a stated purpose. "We already held the data" is not the same as "we may use it for this."
A prompt is a disclosure. Sending personal data to a third-party model provider discloses it to that provider, and depending on where inference runs, may constitute a transfer. Teams pasting customer information into a general-purpose assistant are doing this daily, usually without anyone having assessed it.
Deletion becomes genuinely hard. Removing a record from a database is straightforward. Removing its influence from a trained model is not. This is why controlling what enters training is the meaningful control, rather than promising retrospective removal you cannot perform.
Automated decisions attract scrutiny. Where a system decides something about a person, transparency and explainability expectations apply, and SDAIA's AI governance work points the same direction. A decision you cannot explain is a problem before it is a compliance finding.
Legal counsel interprets the law and files with the regulator. We build the machinery that makes the answers true, and the evidence that demonstrates it.
Finding where personal data actually lives — including the exports, shared drives and shadow databases nobody listed — and building records of processing that reflect reality rather than intention.
Discovery · RoPA · Data mapConsent captured once and honoured everywhere, with purpose scoping, withdrawal that propagates, and an audit trail. Arabic and English interfaces, because the notice must be understood to be valid.
Purpose scoping · WithdrawalWorkflows for access, correction, deletion and portability that actually execute across every system holding the record, within the response window, with evidence of completion.
Access · Deletion · PortabilityRetention schedules enforced by jobs rather than documents, with legal-hold handling and proof that deletion occurred where it was due.
Schedules · Legal hold · ProofDetection capability, severity assessment criteria, and a rehearsed notification path — because 72 hours from awareness is only achievable if awareness comes quickly.
Detection · Rehearsed responseApproved-use registers, data classification for what may enter a model or a prompt, redaction pipelines, and impact assessments for automated decisions.
Approved use · Redaction · DPIAOn scope: Crux is a technology firm, not a law firm. We build and evidence the operational programme — discovery, consent, rights, retention, detection and AI governance. Legal interpretation, regulatory correspondence and formal advice come from your counsel, and everything we build is designed to be demonstrated to them and to the regulator.
Every commitment in a privacy programme depends on knowing where the data is. Programmes that skip discovery end up making undertakings they cannot execute, which is a worse position than an acknowledged gap.
Scanning structured and unstructured stores for personal data, interviewing the people who actually handle it, and producing a map that usually surprises the organisation that commissioned it.
Current state against the obligations, with findings ranked by exposure rather than by ease of fixing. Reviewed jointly with your legal counsel.
Consent architecture, rights workflows, retention automation, detection and logging. The engineering that turns policy statements into system behaviour.
Approved-use register, classification rules for what may enter models and prompts, redaction where needed, and assessment of automated decisions.
Assembling the artefacts a regulator or auditor would request, and rehearsing the breach path against the clock rather than assuming it works.
Periodic re-discovery, notice verification against actual processing, and updates as systems and regulations change. Privacy is a state that decays.
Saudi Arabia's Personal Data Protection Law, supervised by SDAIA. It governs processing of personal data relating to individuals in the Kingdom and reaches organisations outside Saudi Arabia that process such data. Obligations include lawful basis, transparency, data subject rights, security, retention limits and breach notification.
Not on policy documents, which are usually in place. On the operational side: nobody can list every system holding personal data, deletion cannot be executed across all of them, retention exists on paper but nothing enforces it, and consent captured in one channel is not honoured in another.
The position changed with the 2023 amendments and subsequent regulations. Cross-border transfer is permitted subject to conditions, which can include assessing the transfer's impact. It is no longer accurate to describe PDPL as a blanket localisation mandate — though sector regulators may impose stricter rules. Confirm any specific transfer with current SDAIA guidance and your counsel.
Notification to SDAIA is required within 72 hours of becoming aware of a qualifying incident, with affected individuals informed where harm may result. Most organisations cannot meet this because detection takes longer than the deadline — which makes detection the real requirement behind the reporting rule.
The law provides for penalties including significant financial sanctions, with the most serious provisions attaching to unlawful disclosure of sensitive personal data. Assess your exposure with legal counsel against the current published provisions rather than from any vendor's summary, including ours.
Training on personal data is processing and needs a lawful basis and purpose. A prompt containing personal data is a disclosure to the provider and possibly a transfer. Automated decisions raise transparency expectations. Deletion is genuinely hard once data has influenced a trained model, so controlling what enters training is the control that matters.
No. We build the operational and technical programme — discovery, consent, rights workflows, retention automation, breach detection, AI governance and the evidence trail. Legal interpretation and regulatory filings come from your counsel, and everything is built to be demonstrated to them.
This page describes obligations in general terms and is not legal advice. Regulations are updated periodically; confirm current requirements with SDAIA publications and your legal counsel.
A discovery and gap assessment maps personal data across your estate, tests whether a deletion request could actually be executed end to end, and returns findings ranked by exposure — in a form your legal counsel can work from.