ARKHOS | BOARD DECISION BRIEFS
AI governance in Saudi Arabia and the GCC: What should the board approve before scaling?
Decision requested: Approve a pilot with a defined purpose, term and budget, not a general rollout or a commitment extending beyond the pilot. Name one accountable executive, identify who can suspend the pilot immediately, and record who may authorize expansion under the delegation of authority. Management returns with a recommendation to scale, redesign or stop.
Illustrative only, not a client case or actual result: A Saudi group wants an AI assistant to summarize internal policies, then make it available to a UAE subsidiary. Start with a 60-day pilot for named staff and authorized policy documents, with no employee or customer records and no automated decisions. Adding personal records or giving the subsidiary access to conversation logs is a scope change that must be reassessed before allowing it.
Board question before opening subsidiary access: Even if the document set is limited to approved policies, do sign-in records, user prompts or operational logs contain information that could identify someone directly or indirectly? Which fields can the UAE subsidiary, vendor support teams or subprocessors access or receive, and from which countries? Ask Privacy and Legal to document the fields, recipients, their roles and processing and access locations, and assess the applicable safeguards before expanding access [2].
Three decision gates
Before the pilot | Is the scope clear and the decision reversible?
Management records the purpose, current alternative, cost and spending cap. Legal and privacy teams identify the accountable entity, permitted data, where it is processed and stored, and who may access it, including the vendor and subprocessors. Separate pilot conditions from requirements for contracting or scaling.
Before use | Has it been tested in that scope?
Ask for representative Arabic and English task results. Compare factual accuracy, references to source documents and time taken after human review with the baseline. Test document-access restrictions and the designated owner's ability to suspend the tool and return to the usual process. Set success measures and risk limits in advance; an inadequate sample is not evidence of success.
Before expansion | Does the evidence support the decision, and do its conditions still hold?
Compare benefits with full costs, including human verification and support; report Arabic and English results separately so a combined average cannot conceal weaknesses in either. In this example, any unauthorized access or transfer to an unapproved destination pauses the pilot and is escalated immediately to the risk owner. Do not add a new entity, use case or data flow without documented review and approval.
Applicability: SDAIA’s AI Ethics Principles guide responsible governance [1]. Assess Saudi PDPL and its regulations when personal data is processed; where data is transferred or disclosed outside the Kingdom, assess Article 29 and the corresponding regulation [2]. Assess UAE federal scope and exclusions separately, including potentially applicable free-zone or sector rules [3]. NIST AI RMF is voluntary guidance; ISO/IEC 42001 is an AI management-system standard [4] [5]. Neither determines which legal obligations apply. This is a decision aid, not legal advice.
Primary sources
SDAIA AI Ethics Principles: September 2023, v1.0. Saudi PDPL: M/19 (2021), amended by M/148 (2023), Articles 1(4), 1(8), 2 and 29. UAE Federal Decree by Law No. 45 of 2021: Articles 2 and 22–23. NIST AI RMF 1.0: January 2023; ISO/IEC 42001: December 2023.