SAMA expects the banks, payment firms and insurers it supervises to govern technology risk, stay responsible for outsourced services, protect consumers and secure customer data. When those institutions run AI agents, staff overseeing them need training on agent access and incidents, model provider oversight, fair treatment in Arabic and English, and the Personal Data Protection Law.
What does SAMA expect of people who oversee agents?
The Saudi Central Bank has no single rule on AI agents. Its expectations come from frameworks that apply to all technology, and they converge on competent people. In substance:
- Cyber security. SAMA's cyber security framework for regulated entities requires board-level governance of cyber risk, defined roles, security awareness among staff, access control and incident management.
- IT governance and continuity. Institutions must manage IT risk, change and resilience, with clear accountability.
- Outsourcing. An institution that outsources a service, including to a technology provider, remains responsible for it and must oversee the provider, with particular care over data leaving the Kingdom.
- Consumer protection. Customers must be treated fairly and with integrity, given clear information, have their data protected, and have complaints handled properly.
Two national bodies add to this. The Personal Data Protection Law governs personal data processing across sectors. SDAIA, the national data and AI authority, oversees data governance and has published AI ethics principles that stress fairness, privacy, transparency, accountability and human oversight. For a sector view, see what AI agent training organisations in Saudi Arabia need.
Who does it apply to?
Banks, payment service providers, finance companies, insurers and other institutions SAMA licenses and supervises. The PDPL applies more widely, to anyone processing personal data in the Kingdom and, in some cases, to processing outside it about people in Saudi Arabia.
Inside an institution, those overseeing agents include the board and senior management, operations staff releasing agent actions, KYC and onboarding analysts, payments operations staff, contact centre teams, security and IT staff, and vendor managers. KYC is covered in depth in AI agent training for KYC analysts in payments in Saudi Arabia.
What competence do these expectations imply when agents act?
Board and senior management. Leaders must understand which agents run, what each may do at which depth, what is held for people, and how agent incidents are handled. Depth is earned on the institution's own cases, not set by configuration, and agent certificates lapse after 90 days.
Security awareness for agents. Staff need to know how agents are attacked: instructions hidden in documents, pressure in chat, forged evidence. Facts from outside arrive marked external; an external claim cannot authorise a payment. Agent credentials are privileged access and must be scoped, monitored and never shared.
Incident recognition. An agent acting outside its bounds, a sudden run of refusals, or an agent dropped to depth 1 after a certification run that cleared no depth may signal an incident. Staff must stop the agent where needed, preserve the record and escalate.
Provider oversight and data location. Many model providers process data abroad. Vendor managers must know where agent data goes and whether transfer conditions under the PDPL and SAMA's outsourcing expectations are met.
Fair treatment in two languages. Agents may serve customers in Arabic and English. Supervisors must check that disclosures, explanations and complaint responses are accurate and equally clear in both, and that a customer is not disadvantaged by the language they use.
Personal data discipline. Under the PDPL, data must be processed for a clear purpose and limited to what is needed, with people able to exercise their rights. Staff connecting new sources to an agent must check purpose. Staff must recognise when an agent exposes data to the wrong person and report it.
Payments. An agent may prepare a payment or flag one, but releasing money is depth 4 and held for a named person. Payments roles are covered in AI agent training for payments operations staff in Saudi Arabia.
How do institutions meet these expectations and show it?
| Requirement | Control | Evidence it leaves |
|---|---|---|
| Board governance of cyber and IT risk | Board briefing on agents, depth, holds and incidents | Board training records; minutes |
| Security awareness | Agent-specific awareness for staff by role | Training records |
| Access control | Agent credentials scoped and monitored | Access reviews |
| Incident management | Staff trained to stop agents and escalate | Incident records; exercises |
| Outsourcing oversight | Provider assessments, data location records | Vendor files; transfer register |
| Consumer protection | Bilingual QA of agent communications; complaint review | QA results; complaint files |
| PDPL compliance | Agent sources mapped to purposes; breach recognition training | Source register; incident log |
| Competent oversight | Staff assessed before release rights | Assessment results; access grants |
What will SAMA or an auditor ask?
- Which agents do you run, and who oversees each?
- What has the board been trained on?
- Where does agent data go, and on what basis?
- How would staff detect an agent incident?
- How do you check agent communications in both languages?
- Who releases held payments, and how were they trained?
What do these expectations not cover?
They do not replace the PDPL, which applies alongside them, or SDAIA's broader data governance role. They do not cover capital markets activity, which has its own regulator, and they do not prescribe a training curriculum. And they continue to develop, so institutions should track new SAMA publications. Comparable expectations elsewhere include BaFin's in Germany and the CBN's in Nigeria. The guide to AI literacy and certification covers the wider model.
This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.
Where the Academy fits
Pryme Intelligence Academy lessons can be read or listened to as audio, and module tests are set in the learner's own industry and role. Board members take Foundations; operations and KYC staff take the Operator track, whose practicum requires two weeks of real review decisions signed off by a Pryme assessor. Certificates are checkable at source through VerifyDoc.ai.
Questions people ask
Has SAMA issued a rule specifically for AI agents?
Not a single rule. Its cyber security, outsourcing and consumer protection expectations apply to agents, alongside the PDPL and SDAIA's guidance.
Can customer data reach a model provider abroad?
Only where transfer conditions under the PDPL and SAMA's outsourcing expectations are met. Check before deployment.
Should agents serve customers in Arabic?
If customers use Arabic, yes, and supervisors must check that service quality and disclosures are equal across languages.