Skip to content

/articlesGuides

AI governance for enterprises: a practical framework

A practical AI governance framework for companies in Saudi Arabia and Jordan: list your AI systems, name owners, require approvals, protect data, keep records.

Updated
Published
min read
08
On this page
10
On this page10

What AI governance means in practice

Most companies already have a page of AI principles. Governance turns it into five things you can check: a list of AI systems in use, a named owner for each, rules on what each may do, a person who approves risky actions, and records showing all of it happened.

Assistants that connect to your systems make this urgent. An assistant connected to Gmail can send an email; one connected to ServiceNow can open an incident; one connected to Odoo can create a quotation. So part of governance must live inside the product, as permissions and approvals checked before a tool runs.

This guide is a starting point, not legal advice. Confirm your obligations with your lawyer and your data protection officer.

Which laws and frameworks apply in Saudi Arabia and Jordan?

Data protection laws are binding whenever AI processes personal data. AI frameworks and standards are mostly voluntary, but they give you a tested structure and words your auditors already use.

InstrumentIssuerStatus and dateWhat it asks of you
Personal Data Protection Law (PDPL), Saudi ArabiaSDAIA is the enforcing authorityIn force 14 September 2023; fully enforceable from 14 September 2024 after a one-year grace periodLawful processing, data minimisation, records of processing, rules on transfers outside the Kingdom, a data protection officer in defined cases
AI Ethics PrinciplesSaudi Data and Artificial Intelligence Authority (SDAIA)Version 1.0 published September 2023Seven principles: fairness; privacy and security; humanity; social and environmental benefits; reliability and safety; transparency and explainability; accountability and responsibility
Personal Data Protection Law No. 24 of 2023, JordanMinistry of Digital Economy and Entrepreneurship, with a Personal Data Protection BoardPublished 17 September 2023; in effect 17 March 2024, with one year for existing processing to alignPrior consent unless an exception applies, data subject rights, limits on transfers abroad, breach notification, a data protection officer in defined cases
AI Risk Management Framework (AI RMF 1.0)US National Institute of Standards and Technology (NIST)Published 26 January 2023; voluntaryFour functions: Govern, Map, Measure, Manage
ISO/IEC 42001:2023ISO and IECPublished December 2023Requirements for an AI management system (AIMS) run on Plan-Do-Check-Act; certification is voluntary and done by independent certification bodies

Step 1: List your AI systems and rate the risk

You cannot govern what you have not listed. The Govern function of the NIST AI RMF calls for an inventory of AI systems. For each one, record its purpose, owner, the data it reads and the systems it can change. Include tools that single teams bought on their own.

Then rate risk by what the system can do, not by how clever the model is. For assistants, rate each action, because one assistant can hold a harmless search tool and a risky send tool.

Action classExamplesSuggested default
ReadSearch logs, read an incident, read an emailAllowed for the right roles; logged
WriteCreate a draft, add a label, update a ticket fieldAllowed or approval, by your rules
DestructiveDelete or trash data, close recordsApproval; limited roles
External sendSend or forward an email, send a quotation to a customerApproval of the exact content

Step 2: Name the owners

SDAIA’s seventh principle, accountability and responsibility, holds those who design, sell, buy, build, own and assess AI systems responsible for their effects, with human oversight across the system’s life. NIST also asks for documented roles and for senior leaders to own AI risk decisions. A short list of named roles covers most of this.

  • Executive sponsor: sets how much AI risk the company accepts and signs off high-risk uses.
  • AI review group: security, legal, data protection, IT and business owners; approves new systems and reviews incidents.
  • Data protection officer: required in defined cases under both the Saudi and the Jordanian law; reviews personal data flows before launch.
  • System owner: one person accountable for one assistant, its settings, its rules and its audit review.
  • Approvers: named people who may approve specific actions.
  • Internal audit: tests the controls against the audit log, not against interviews.

Step 3: Design approvals people will actually use

Human oversight fails in two ways: nobody is asked, or everyone is asked so often that approving becomes a reflex.

Put the approval on the action, not on the conversation. Reading data rarely needs a person. Sending, deleting or changing records outside the platform usually does.

A good approval shows the exact action and content, such as the email text, says who asked, and is as easy to reject as to accept. The task waits until someone decides, and the decision is recorded.

Review approvals every month. If an action is always approved unchanged, consider letting it run. If approvers keep rejecting the same kind of request, fix the assistant’s instructions.

Step 4: Apply data protection to AI work

An assistant moves personal data in new ways: into prompts, through model providers, into reports and into memory. Map each flow before launch.

The Saudi PDPL stresses data minimisation and records of processing. SDAIA’s rules on transfers outside the Kingdom rely on safeguards such as standard contract clauses where the destination has not been found adequate.

Jordan’s law bases processing on prior consent unless an exception applies, limits transfers to places with weaker protection, and requires breach notice to the people affected and the regulator within the deadlines it sets.

Ask every AI vendor where prompts, tool results and logs are processed and stored. Get the answer in writing.

  • Send the model only the fields a task needs.
  • Add AI processing, including model providers, to your records of processing.
  • Decide what the assistant may remember, for how long and about whom.
  • Apply retention rules to transcripts and generated reports, not only to source data.

Step 5: Audit, monitor and handle incidents

For any task, your records should answer: who asked, what the assistant did, with which inputs, who approved it and what came back. Chat transcripts alone do not answer that.

NIST’s Measure and Manage functions add the routine: watch behaviour in production, track new risks, and plan how to respond, recover and retire a system.

Treat AI incidents like security incidents: a wrong email sent, a record changed without authority, personal data in a report. Decide who triages, how to pause an assistant or disconnect a tool, and how the lesson changes your rules.

What Botify enforces in the product

Written rules work only if the software enforces them. In Botify, every call to a connected system passes the same steps: permission check, your rules, approval if needed, then the system.

  • Roles: owner, admin, operator, member and viewer. Members and above decide approvals; viewers only look.
  • Every tool has a risk class (read, write, destructive, external send) and an approval mode: never, by your rules, or always. Gmail send and ServiceNow incident creation are always approved by a person.
  • Your rules can allow, deny or require approval. A deny wins, and rules can add approvals but cannot remove an “always”.
  • Quotations need approval only when you turn it on, for the assistant or through a rule.
  • Approvals are answered with buttons in WhatsApp and Telegram, and in the Inbox for everything else. Each approval stores who decided and when.
  • The audit log is in the console under Insights.

What a platform does not do for you

Controls like these support a governance programme; they do not replace it. Using any platform does not by itself make a company compliant with the PDPL, Jordan’s law or any standard.

Botify does not state certifications or a hosting region on this site. Ask for the details your review needs. The framework above, owned by named people and reviewed on a schedule, does the real work.

Common questions

01

What is an AI governance framework?

It is the set of rules, roles, controls and reviews a company uses to decide which AI systems it runs and how they are checked. It covers inventory, risk, ownership, human approval, data protection and audit. NIST AI RMF and ISO/IEC 42001 are common reference structures.

02

Does Saudi Arabia’s PDPL apply to AI systems?

Yes, whenever an AI system processes personal data. The PDPL has been fully enforceable since 14 September 2024 and also reaches organisations outside the Kingdom that process the personal data of its residents.

03

Is ISO/IEC 42001 certification required in Saudi Arabia or Jordan?

ISO describes certification against ISO/IEC 42001 as voluntary, done by independent bodies, and says the standard does not replace laws. Check your sector regulator and key contracts for specific requirements.

04

How is AI governance different from data governance?

Data governance controls the quality, access and life of data. AI governance adds control over what automated systems do with it: which actions they may take, when a person must approve, and how results are monitored.

05

Which AI actions should always need human approval?

Actions that leave the company or are hard to undo: sending or forwarding email, sending documents to customers, deleting data, and changing records others rely on. Read-only lookups can usually run directly.

06

Who should own AI governance in a company?

A named executive should own the risk appetite, supported by security, legal, data protection, IT and the business. Each assistant also needs its own owner who keeps its rules current and reviews its audit log.

Sources

  1. [1]Saudi Arabia’s Personal Data Protection Law becomes enforceable (Clyde & Co, 2024)
  2. [2]SDAIA AI Ethics Principles, Version 1.0 (September 2023)
  3. [3]SDAIA AI Ethics Principles (sdaia.gov.sa)
  4. [4]Jordan issues first personal data protection law (Clyde & Co, 2023)
  5. [5]NIST AI Risk Management Framework (AI RMF 1.0)
  6. [6]NIST AI RMF Core: Govern, Map, Measure, Manage
  7. [7]ISO/IEC 42001:2023 AI management systems
  8. [8]ISO/IEC 42001 explained (ISO)

Where you’ll find this in Botify

Start with one assistant on WhatsApp or your website.

Book a demo and we’ll show you an assistant answering questions like your customers’. We set up the first one with you and review the results together before you add more.