One region. Several supervisory perimeters. Two real approval gates that should not be guessed away.
Oonagh van den Berg | RAW Compliance
A bank can have one AI model and still face different questions at each border. Does the regulator expect a control record, an approval before a contract, a model validation, or a bounded test? The Middle East is a good place to see why "AI policy" is not a single regulatory status.
Issue 2 mapped APAC, where consultation, sandbox, guidance and existing duties had to remain separate. Here the separation is sharper. Qatar's central bank names approval points for certain AI activity. Israel's bank-model directive expressly includes AI models. The UAE has more than one financial-services jurisdiction, each with its own supervisory text. Elsewhere a fintech sandbox or cyber framework cannot be turned into a general licence for AI production.
UAE onshore - consumer protection and the model inventory
The Central Bank of the UAE issued a consumer-protection Guidance Note on responsible AI and machine learning for licensed financial institutions, including insurers. It expects governance proportionate to the firm, board and senior-management accountability, an inventory with purpose and risk rating, testing for bias after material change, meaningful human oversight and the ability to cease use. It also points to the CBUAE Model Management Standards and existing outsourcing and consumer-protection obligations. The note supplements, rather than replaces, binding requirements. For a vendor, the hard question is not whether a model can produce an explanation. It is whether the institution can join that explanation to its own inventory, approval, vendor and monitoring record.
DIFC and ADGM - do not put free-zone firms into an onshore bucket
The DFSA has issued a 2026 letter on AI risk-management expectations for firms in the DIFC. Its public listing establishes that supervisory signal; this map does not claim a detailed clause from a letter not reproduced on the accessible page. The ADGM FSRA rulebook's Big Data Analytics and AI section is more directly readable: governance, governing-body accountability, reliable and explainable material models, review before deployment and periodically after, fairness and customer disclosure. Those requirements must be mapped to the right licensed entity. A group policy copied across the UAE is a starting control, not evidence that three supervisors asked the same question.
Sources: https://www.dfsa.ae/your-resources/publications-reports/seo-letters-1/2026/dfsa-regulatory-expectations-artificial-intelligence-risk-management-difc ; https://en.adgm.thomsonreuters.com/rulebook/big-data-analytics-and-artificial-intelligence-ai
Qatar - approval is not one blanket checkbox
Qatar Central Bank's 2024 AI Guideline entered into force on 4 September 2024. Its section 11 distinguishes at least two approval triggers: an entity needs official QCB approval before launching a new AI system as a Provider or materially modifying an existing one; it also needs approval before signing a high-risk AI purchase, licensing or outsourcing agreement. QCB may direct a particular system to a sandbox before approval. The register and risk-assessment duties sit alongside those gates. Do not translate this into "every AI deployment needs pre-approval" or "sandbox entry approves production". A product team must know which role the entity plays and whether a proposed contract is high-risk before it sets a release date.
Source (QCB's scanned guideline, section 11): https://www.qcb.gov.qa/Services/Financial%20Technology/QCB_Artificial_Intelligence_Guideline.pdf
Jordan - an AI framework built for banks
Jordan's central bank published its AI Framework for the Banking Sector in July 2025. Its scope covers CBJ-regulated banking institutions using AI they develop, procure or outsource. It sets out governance, risk classification, ethics, explainability and customer autonomy, data management, model testing, incident response and third-party risk. This is a sector-specific framework, not a generic national AI aspiration. The practical implementation unit is a use case with a risk owner, evidence of testing, an override path and a vendor record. The board cannot delegate the decision trail to the model provider.
Source: https://www.cbj.gov.jo/EBV4.0/Root_Storage/AR/FinCert/Ai-Framework_English_Version.pdf
Israel - an effective model-risk directive reaches AI
The Bank of Israel's Proper Conduct of Banking Business Directive 369 on model-risk management expressly says it applies to AI models. Its August 2024 circular states an August 2025 effective date, with phased processes for models already in use. The directive brings model inventories, development and validation, governance and supplier models into the same risk frame. This is not merely a discussion paper about AI possibilities. If an AI system supports a credit, fraud or customer decision, the question is which model version was validated, who challenged it and how a supplier change flows through the institution's controls.
Sources: https://www.boi.org.il/media/roajuh3i/h2792_en.pdf ; https://www.boi.org.il/media/px3dfkjk/369_en.pdf
Saudi Arabia - a mandatory fraud-control spine, not a general AI licence
SAMA's Counter-Fraud Framework is mandatory for member organisations it notifies and gives control requirements for fraud governance, prevention, detection and response. It is not, by itself, a standalone AI deployment code. A bank using AI in detection still has to show that model alerts connect to the framework's controls, escalation and management evidence. "AI found a pattern" is a signal; it is not the end of the investigation or a waiver from existing duties. One national layer sits above it. On 29 September 2026 the Saudi Press Agency reported that the Saudi Data and AI Authority (SDAIA) has a National AI Risk Management Framework: a unified methodology for identifying, assessing, addressing and monitoring AI risk across government and private-sector entities. This map has not found a SAMA instrument that makes that framework a binding bank requirement, so treat it as the national reference for risk method and confirm its status for your institution with SAMA before relying on it.
Sources: https://rulebook.sama.gov.sa/en/entiresection/2217 ; https://www.spa.gov.sa/en/N2688337 ; https://www.arabnews.com/business/saudi-arabia-strengthens-ai-governance-with-unified-risk-framework-3003753
Bahrain - the sandbox boundary
The Central Bank of Bahrain's fintech sandbox provides a supervised test route for licensees and innovators. Entry requires CBB authorisation; AML/CFT, customer-money and confidentiality safeguards still apply. It is a testing lane, not a blanket AI regulation or an automatic production permission. A firm should label trial data, duration, customer protections and exit conditions before a pilot becomes a live control.
Source: https://www.cbb.gov.bh/fintech/
Kuwait and Oman - existing controls while AI-specific questions mature
Kuwait's central bank has a Cyber and Operational Resilience Framework and a Wolooj fintech sandbox. Oman lists its banking regulatory framework and fintech policies, including cybersecurity and resilience. The primary materials checked for this issue do not establish a dedicated bank-AI approval rule in either jurisdiction. That is a scoped research finding, not proof no other circular exists. For a regional rollout, carry existing cyber, data, outsourcing and consumer duties into the AI decision and confirm any institution-specific direction with the supervisor before production.
Sources: https://www.cbk.gov.kw/en/supervision/cyber_operational_resilience_framework ; https://www.cbk.gov.kw/en/legislation-and-regulation/innovation-hub/innovation-framework ; https://cbo.gov.om/Pages/RegulatoryFramework.aspx ; https://cbo.gov.om/fintechkipolicies
Turkey, Lebanon and Iraq - no confirmed bank AI rule
Turkey: Turkish business press reported in June 2026 that the banking regulator BDDK is preparing a framework for AI in banks, built around data security, transparency and technological independence. The report itself says there was no official BDDK statement, so this map treats it as a signal, not a rule. Existing BDDK rules on bank information systems and electronic banking still apply. Lebanon and Iraq: the sources checked show no AI-specific bank instrument from Banque du Liban or the Central Bank of Iraq. The Central Bank of Iraq has a multi-year banking reform and digital-bank licensing programme, which is not an AI rule. These are scoped findings, not proof that nothing exists.
Sources: https://www.bdturkey.com/bankacilikta-yapay-zeka-duzenlemesi-yolda ; https://www.bddk.gov.tr/Mevzuat/DokumanGetir/1063 ; https://www.bdl.gov.lb/intermediatecirculars.php?page=1 ; https://cbi.iq/static/uploads/up/file-176050867224562.pdf
The operating model that travels
Keep one evidence spine - a risk-ranked use-case inventory, model and data provenance, accountable owner, independent challenge, customer impact, human authority, vendor assurance, monitoring and exit. Add local gates as fields, not footnotes. QCB provider and high-risk contract approvals are different events. Israeli validation is not a Qatar authorisation. A Bahrain sandbox is not an onshore UAE policy. The evidence pack must say which entity, system, contract and customer journey it covers.
The trap is a regional dashboard that turns every market green because each has a link to a "policy". A live directive, a guidance note, a sandbox and a supervisory letter are not the same kind of instrument. If the team cannot name the next gate and the person who owns it, it has a map of documents, not a route to production.
Next issue: Europe.
Source and status note
Positions checked against regulator-primary materials on 25 September 2026 and re-checked for changes on 10 October 2026. The one change found is the SDAIA National AI Risk Management Framework report of 29 September (Saudi Press Agency). No change was found for the UAE, Qatar, Jordan, Israel, Bahrain, Kuwait or Oman. The DFSA letter’s detailed terms and any institution-specific directions require direct confirmation before relying on them. This is a market map, not legal advice. Turkey, Lebanon and Iraq were added on 11 October 2026 as scoped findings. The Turkish item rests on press reporting with no BDDK statement.
THE AI REGULATION MARKET MAP • MIDDLE EAST • RAW COMPLIANCE
One region. Several supervisory perimeters. Two real approval gates that should not be guessed away.
Oonagh van den Berg | RAW Compliance
A bank can have one AI model and still face different questions at each border. Does the regulator expect a control record, an approval before a contract, a model validation, or a bounded test? The Middle East is a good place to see why "AI policy" is not a single regulatory status.
Issue 2 mapped APAC, where consultation, sandbox, guidance and existing duties had to remain separate. Here the separation is sharper. Qatar's central bank names approval points for certain AI activity. Israel's bank-model directive expressly includes AI models. The UAE has more than one financial-services jurisdiction, each with its own supervisory text. Elsewhere a fintech sandbox or cyber framework cannot be turned into a general licence for AI production.
UAE onshore - consumer protection and the model inventory
The Central Bank of the UAE issued a consumer-protection Guidance Note on responsible AI and machine learning for licensed financial institutions, including insurers. It expects governance proportionate to the firm, board and senior-management accountability, an inventory with purpose and risk rating, testing for bias after material change, meaningful human oversight and the ability to cease use. It also points to the CBUAE Model Management Standards and existing outsourcing and consumer-protection obligations. The note supplements, rather than replaces, binding requirements. For a vendor, the hard question is not whether a model can produce an explanation. It is whether the institution can join that explanation to its own inventory, approval, vendor and monitoring record.
DIFC and ADGM - do not put free-zone firms into an onshore bucket
The DFSA has issued a 2026 letter on AI risk-management expectations for firms in the DIFC. Its public listing establishes that supervisory signal; this map does not claim a detailed clause from a letter not reproduced on the accessible page. The ADGM FSRA rulebook's Big Data Analytics and AI section is more directly readable: governance, governing-body accountability, reliable and explainable material models, review before deployment and periodically after, fairness and customer disclosure. Those requirements must be mapped to the right licensed entity. A group policy copied across the UAE is a starting control, not evidence that three supervisors asked the same question.
Sources: https://www.dfsa.ae/your-resources/publications-reports/seo-letters-1/2026/dfsa-regulatory-expectations-artificial-intelligence-risk-management-difc ; https://en.adgm.thomsonreuters.com/rulebook/big-data-analytics-and-artificial-intelligence-ai
Qatar - approval is not one blanket checkbox
Qatar Central Bank's 2024 AI Guideline entered into force on 4 September 2024. Its section 11 distinguishes at least two approval triggers: an entity needs official QCB approval before launching a new AI system as a Provider or materially modifying an existing one; it also needs approval before signing a high-risk AI purchase, licensing or outsourcing agreement. QCB may direct a particular system to a sandbox before approval. The register and risk-assessment duties sit alongside those gates. Do not translate this into "every AI deployment needs pre-approval" or "sandbox entry approves production". A product team must know which role the entity plays and whether a proposed contract is high-risk before it sets a release date.
Source (QCB's scanned guideline, section 11): https://www.qcb.gov.qa/Services/Financial%20Technology/QCB_Artificial_Intelligence_Guideline.pdf
Jordan - an AI framework built for banks
Jordan's central bank published its AI Framework for the Banking Sector in July 2025. Its scope covers CBJ-regulated banking institutions using AI they develop, procure or outsource. It sets out governance, risk classification, ethics, explainability and customer autonomy, data management, model testing, incident response and third-party risk. This is a sector-specific framework, not a generic national AI aspiration. The practical implementation unit is a use case with a risk owner, evidence of testing, an override path and a vendor record. The board cannot delegate the decision trail to the model provider.
Source: https://www.cbj.gov.jo/EBV4.0/Root_Storage/AR/FinCert/Ai-Framework_English_Version.pdf
Israel - an effective model-risk directive reaches AI
The Bank of Israel's Proper Conduct of Banking Business Directive 369 on model-risk management expressly says it applies to AI models. Its August 2024 circular states an August 2025 effective date, with phased processes for models already in use. The directive brings model inventories, development and validation, governance and supplier models into the same risk frame. This is not merely a discussion paper about AI possibilities. If an AI system supports a credit, fraud or customer decision, the question is which model version was validated, who challenged it and how a supplier change flows through the institution's controls.
Sources: https://www.boi.org.il/media/roajuh3i/h2792_en.pdf ; https://www.boi.org.il/media/px3dfkjk/369_en.pdf
Saudi Arabia - a mandatory fraud-control spine, not a general AI licence
SAMA's Counter-Fraud Framework is mandatory for member organisations it notifies and gives control requirements for fraud governance, prevention, detection and response. It is not, by itself, a standalone AI deployment code. A bank using AI in detection still has to show that model alerts connect to the framework's controls, escalation and management evidence. "AI found a pattern" is a signal; it is not the end of the investigation or a waiver from existing duties. One national layer sits above it. On 29 September 2026 the Saudi Press Agency reported that the Saudi Data and AI Authority (SDAIA) has a National AI Risk Management Framework: a unified methodology for identifying, assessing, addressing and monitoring AI risk across government and private-sector entities. This map has not found a SAMA instrument that makes that framework a binding bank requirement, so treat it as the national reference for risk method and confirm its status for your institution with SAMA before relying on it.
Sources: https://rulebook.sama.gov.sa/en/entiresection/2217 ; https://www.spa.gov.sa/en/N2688337 ; https://www.arabnews.com/business/saudi-arabia-strengthens-ai-governance-with-unified-risk-framework-3003753
Bahrain - the sandbox boundary
The Central Bank of Bahrain's fintech sandbox provides a supervised test route for licensees and innovators. Entry requires CBB authorisation; AML/CFT, customer-money and confidentiality safeguards still apply. It is a testing lane, not a blanket AI regulation or an automatic production permission. A firm should label trial data, duration, customer protections and exit conditions before a pilot becomes a live control.
Source: https://www.cbb.gov.bh/fintech/
Kuwait and Oman - existing controls while AI-specific questions mature
Kuwait's central bank has a Cyber and Operational Resilience Framework and a Wolooj fintech sandbox. Oman lists its banking regulatory framework and fintech policies, including cybersecurity and resilience. The primary materials checked for this issue do not establish a dedicated bank-AI approval rule in either jurisdiction. That is a scoped research finding, not proof no other circular exists. For a regional rollout, carry existing cyber, data, outsourcing and consumer duties into the AI decision and confirm any institution-specific direction with the supervisor before production.
Sources: https://www.cbk.gov.kw/en/supervision/cyber_operational_resilience_framework ; https://www.cbk.gov.kw/en/legislation-and-regulation/innovation-hub/innovation-framework ; https://cbo.gov.om/Pages/RegulatoryFramework.aspx ; https://cbo.gov.om/fintechkipolicies
Turkey, Lebanon and Iraq - no confirmed bank AI rule
Turkey: Turkish business press reported in June 2026 that the banking regulator BDDK is preparing a framework for AI in banks, built around data security, transparency and technological independence. The report itself says there was no official BDDK statement, so this map treats it as a signal, not a rule. Existing BDDK rules on bank information systems and electronic banking still apply. Lebanon and Iraq: the sources checked show no AI-specific bank instrument from Banque du Liban or the Central Bank of Iraq. The Central Bank of Iraq has a multi-year banking reform and digital-bank licensing programme, which is not an AI rule. These are scoped findings, not proof that nothing exists.
Sources: https://www.bdturkey.com/bankacilikta-yapay-zeka-duzenlemesi-yolda ; https://www.bddk.gov.tr/Mevzuat/DokumanGetir/1063 ; https://www.bdl.gov.lb/intermediatecirculars.php?page=1 ; https://cbi.iq/static/uploads/up/file-176050867224562.pdf
The operating model that travels
Keep one evidence spine - a risk-ranked use-case inventory, model and data provenance, accountable owner, independent challenge, customer impact, human authority, vendor assurance, monitoring and exit. Add local gates as fields, not footnotes. QCB provider and high-risk contract approvals are different events. Israeli validation is not a Qatar authorisation. A Bahrain sandbox is not an onshore UAE policy. The evidence pack must say which entity, system, contract and customer journey it covers.
The trap is a regional dashboard that turns every market green because each has a link to a "policy". A live directive, a guidance note, a sandbox and a supervisory letter are not the same kind of instrument. If the team cannot name the next gate and the person who owns it, it has a map of documents, not a route to production.
Next issue: Europe.
Source and status note
Positions checked against regulator-primary materials on 25 September 2026 and re-checked for changes on 10 October 2026. The one change found is the SDAIA National AI Risk Management Framework report of 29 September (Saudi Press Agency). No change was found for the UAE, Qatar, Jordan, Israel, Bahrain, Kuwait or Oman. The DFSA letter’s detailed terms and any institution-specific directions require direct confirmation before relying on them. This is a market map, not legal advice. Turkey, Lebanon and Iraq were added on 11 October 2026 as scoped findings. The Turkish item rests on press reporting with no BDDK statement.
THE AI REGULATION MARKET MAP • MIDDLE EAST • RAW COMPLIANCE
One region. Several supervisory perimeters. Two real approval gates that should not be guessed away.
Oonagh van den Berg | RAW Compliance
A bank can have one AI model and still face different questions at each border. Does the regulator expect a control record, an approval before a contract, a model validation, or a bounded test? The Middle East is a good place to see why "AI policy" is not a single regulatory status.
Issue 2 mapped APAC, where consultation, sandbox, guidance and existing duties had to remain separate. Here the separation is sharper. Qatar's central bank names approval points for certain AI activity. Israel's bank-model directive expressly includes AI models. The UAE has more than one financial-services jurisdiction, each with its own supervisory text. Elsewhere a fintech sandbox or cyber framework cannot be turned into a general licence for AI production.
UAE onshore - consumer protection and the model inventory
The Central Bank of the UAE issued a consumer-protection Guidance Note on responsible AI and machine learning for licensed financial institutions, including insurers. It expects governance proportionate to the firm, board and senior-management accountability, an inventory with purpose and risk rating, testing for bias after material change, meaningful human oversight and the ability to cease use. It also points to the CBUAE Model Management Standards and existing outsourcing and consumer-protection obligations. The note supplements, rather than replaces, binding requirements. For a vendor, the hard question is not whether a model can produce an explanation. It is whether the institution can join that explanation to its own inventory, approval, vendor and monitoring record.
DIFC and ADGM - do not put free-zone firms into an onshore bucket
The DFSA has issued a 2026 letter on AI risk-management expectations for firms in the DIFC. Its public listing establishes that supervisory signal; this map does not claim a detailed clause from a letter not reproduced on the accessible page. The ADGM FSRA rulebook's Big Data Analytics and AI section is more directly readable: governance, governing-body accountability, reliable and explainable material models, review before deployment and periodically after, fairness and customer disclosure. Those requirements must be mapped to the right licensed entity. A group policy copied across the UAE is a starting control, not evidence that three supervisors asked the same question.
Sources: https://www.dfsa.ae/your-resources/publications-reports/seo-letters-1/2026/dfsa-regulatory-expectations-artificial-intelligence-risk-management-difc ; https://en.adgm.thomsonreuters.com/rulebook/big-data-analytics-and-artificial-intelligence-ai
Qatar - approval is not one blanket checkbox
Qatar Central Bank's 2024 AI Guideline entered into force on 4 September 2024. Its section 11 distinguishes at least two approval triggers: an entity needs official QCB approval before launching a new AI system as a Provider or materially modifying an existing one; it also needs approval before signing a high-risk AI purchase, licensing or outsourcing agreement. QCB may direct a particular system to a sandbox before approval. The register and risk-assessment duties sit alongside those gates. Do not translate this into "every AI deployment needs pre-approval" or "sandbox entry approves production". A product team must know which role the entity plays and whether a proposed contract is high-risk before it sets a release date.
Source (QCB's scanned guideline, section 11): https://www.qcb.gov.qa/Services/Financial%20Technology/QCB_Artificial_Intelligence_Guideline.pdf
Jordan - an AI framework built for banks
Jordan's central bank published its AI Framework for the Banking Sector in July 2025. Its scope covers CBJ-regulated banking institutions using AI they develop, procure or outsource. It sets out governance, risk classification, ethics, explainability and customer autonomy, data management, model testing, incident response and third-party risk. This is a sector-specific framework, not a generic national AI aspiration. The practical implementation unit is a use case with a risk owner, evidence of testing, an override path and a vendor record. The board cannot delegate the decision trail to the model provider.
Source: https://www.cbj.gov.jo/EBV4.0/Root_Storage/AR/FinCert/Ai-Framework_English_Version.pdf
Israel - an effective model-risk directive reaches AI
The Bank of Israel's Proper Conduct of Banking Business Directive 369 on model-risk management expressly says it applies to AI models. Its August 2024 circular states an August 2025 effective date, with phased processes for models already in use. The directive brings model inventories, development and validation, governance and supplier models into the same risk frame. This is not merely a discussion paper about AI possibilities. If an AI system supports a credit, fraud or customer decision, the question is which model version was validated, who challenged it and how a supplier change flows through the institution's controls.
Sources: https://www.boi.org.il/media/roajuh3i/h2792_en.pdf ; https://www.boi.org.il/media/px3dfkjk/369_en.pdf
Saudi Arabia - a mandatory fraud-control spine, not a general AI licence
SAMA's Counter-Fraud Framework is mandatory for member organisations it notifies and gives control requirements for fraud governance, prevention, detection and response. It is not, by itself, a standalone AI deployment code. A bank using AI in detection still has to show that model alerts connect to the framework's controls, escalation and management evidence. "AI found a pattern" is a signal; it is not the end of the investigation or a waiver from existing duties. One national layer sits above it. On 29 September 2026 the Saudi Press Agency reported that the Saudi Data and AI Authority (SDAIA) has a National AI Risk Management Framework: a unified methodology for identifying, assessing, addressing and monitoring AI risk across government and private-sector entities. This map has not found a SAMA instrument that makes that framework a binding bank requirement, so treat it as the national reference for risk method and confirm its status for your institution with SAMA before relying on it.
Sources: https://rulebook.sama.gov.sa/en/entiresection/2217 ; https://www.spa.gov.sa/en/N2688337 ; https://www.arabnews.com/business/saudi-arabia-strengthens-ai-governance-with-unified-risk-framework-3003753
Bahrain - the sandbox boundary
The Central Bank of Bahrain's fintech sandbox provides a supervised test route for licensees and innovators. Entry requires CBB authorisation; AML/CFT, customer-money and confidentiality safeguards still apply. It is a testing lane, not a blanket AI regulation or an automatic production permission. A firm should label trial data, duration, customer protections and exit conditions before a pilot becomes a live control.
Source: https://www.cbb.gov.bh/fintech/
Kuwait and Oman - existing controls while AI-specific questions mature
Kuwait's central bank has a Cyber and Operational Resilience Framework and a Wolooj fintech sandbox. Oman lists its banking regulatory framework and fintech policies, including cybersecurity and resilience. The primary materials checked for this issue do not establish a dedicated bank-AI approval rule in either jurisdiction. That is a scoped research finding, not proof no other circular exists. For a regional rollout, carry existing cyber, data, outsourcing and consumer duties into the AI decision and confirm any institution-specific direction with the supervisor before production.
Sources: https://www.cbk.gov.kw/en/supervision/cyber_operational_resilience_framework ; https://www.cbk.gov.kw/en/legislation-and-regulation/innovation-hub/innovation-framework ; https://cbo.gov.om/Pages/RegulatoryFramework.aspx ; https://cbo.gov.om/fintechkipolicies
Turkey, Lebanon and Iraq - no confirmed bank AI rule
Turkey: Turkish business press reported in June 2026 that the banking regulator BDDK is preparing a framework for AI in banks, built around data security, transparency and technological independence. The report itself says there was no official BDDK statement, so this map treats it as a signal, not a rule. Existing BDDK rules on bank information systems and electronic banking still apply. Lebanon and Iraq: the sources checked show no AI-specific bank instrument from Banque du Liban or the Central Bank of Iraq. The Central Bank of Iraq has a multi-year banking reform and digital-bank licensing programme, which is not an AI rule. These are scoped findings, not proof that nothing exists.
Sources: https://www.bdturkey.com/bankacilikta-yapay-zeka-duzenlemesi-yolda ; https://www.bddk.gov.tr/Mevzuat/DokumanGetir/1063 ; https://www.bdl.gov.lb/intermediatecirculars.php?page=1 ; https://cbi.iq/static/uploads/up/file-176050867224562.pdf
The operating model that travels
Keep one evidence spine - a risk-ranked use-case inventory, model and data provenance, accountable owner, independent challenge, customer impact, human authority, vendor assurance, monitoring and exit. Add local gates as fields, not footnotes. QCB provider and high-risk contract approvals are different events. Israeli validation is not a Qatar authorisation. A Bahrain sandbox is not an onshore UAE policy. The evidence pack must say which entity, system, contract and customer journey it covers.
The trap is a regional dashboard that turns every market green because each has a link to a "policy". A live directive, a guidance note, a sandbox and a supervisory letter are not the same kind of instrument. If the team cannot name the next gate and the person who owns it, it has a map of documents, not a route to production.
Next issue: Europe.
Source and status note
Positions checked against regulator-primary materials on 25 September 2026 and re-checked for changes on 10 October 2026. The one change found is the SDAIA National AI Risk Management Framework report of 29 September (Saudi Press Agency). No change was found for the UAE, Qatar, Jordan, Israel, Bahrain, Kuwait or Oman. The DFSA letter’s detailed terms and any institution-specific directions require direct confirmation before relying on them. This is a market map, not legal advice. Turkey, Lebanon and Iraq were added on 11 October 2026 as scoped findings. The Turkish item rests on press reporting with no BDDK statement.
THE AI REGULATION MARKET MAP • MIDDLE EAST • RAW COMPLIANCE
