Which Saudi Arabian rules apply to AI agents?

Data protection first, then sector supervision. The Personal Data Protection Law governs any agent processing personal data in the Kingdom, with SDAIA, the national data and AI authority, as its overseer and the source of national AI ethics guidance. For banks, insurers and finance companies, the Saudi Central Bank (SAMA) adds expectations on governance, cybersecurity, outsourcing and customer protection.

Who regulates what?

What do the rules require, in substance?

The Personal Data Protection Law requires controllers to process personal data only for a specific, clear and lawful purpose, to collect no more than needed, to keep it accurate and protected, and to respect people's rights to be informed, to access, to correct and to have data destroyed when no longer needed. Consent plays a central role, with defined exceptions. Transfers outside the Kingdom are restricted and allowed only under the conditions the law and its regulations set. Breaches must be notified. Processors, such as an agent platform or model provider, must be chosen with care and bound to protect the data.

SDAIA's AI guidance sets out principles for responsible AI, such as fairness, transparency, accountability and human oversight. It informs how supervisors and customers expect agents to be governed, even where it is not itself a binding rule.

SAMA expects regulated entities to govern technology risk at board level, meet its cybersecurity expectations, manage outsourcing so the entity stays accountable and the supervisor keeps access, and treat customers fairly with effective complaint handling.

This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.

Where does the line fall, and why?

Two Saudi-specific points shape it. Data location matters: because transfers outside the Kingdom are restricted, the choice of where an agent's model runs is a legal question, not just a technical one. And SAMA's cybersecurity and outsourcing expectations mean an agent with access to core systems is treated as a privileged identity under close control.

• Agent may: answer customer questions in Arabic and English from approved content, gather documents for onboarding, monitor transactions, reconcile payments, draft complaint responses.
• Agent may, within limits: process routine service requests after authentication, reverse small charges, schedule confirmed payments to known beneficiaries.
• Stays with a person: payments to new beneficiaries or above limits, credit refusals, account restrictions, decisions on suspected fraud, any new use of personal data, and any transfer of data outside approved arrangements.

An illustrative example

Take a Saudi finance company that plans to deploy an agent for instalment finance enquiries. The vendor's model runs in a data centre abroad. The company's data protection officer flags that customer messages contain personal data and possibly financial details, so sending them to the model is a transfer outside the Kingdom.

The company has three options: use a deployment hosted in the Kingdom, strip personal data before it reaches the model, or establish a lawful basis for the transfer under the law's conditions. It chooses in-Kingdom hosting for customer-facing work. It then registers the agent in its SAMA outsourcing records, scopes its access to the finance system to read-only for enquiries, and holds any change to a customer's repayment plan for a collections officer.

Which controls meet the rules?

• Data location decisions recorded for every agent and model.
• Purpose-bound sources and minimised inputs.
• Processor agreements with platforms and model providers.
• Outsourcing and cybersecurity assessments for SAMA-regulated firms.
• Holds on payments, credit refusals and account restrictions.
• Records that let the firm answer data subject requests and supervisor questions.

For comparison, see which Indian rules apply to AI agents in banking and which Kenyan rules apply to AI agents. Groups with UK operations will also meet UK operational resilience rules for AI agents. The full map is in AI agent regulation, mapped to controls.

Where Pryme fits

Pryme's minimum chain checks every action against role scope, certificate, sources, the customer's own limits and the platform guardrails. A Saudi firm can set its own limits so that payments to new beneficiaries are always held, and each verdict is filed with the bound that decided it.