Siam Thanat Hack Co., Ltd.

Thai AI Governance and Sectoral AI Risk Guidelines: BOT, SEC, OIC, ETDA, NCSA, and Technical AI Assurance

A unified reference for Thai sectoral AI governance across regulators and public agencies, integrating BOT circular Wor 5994/2568, the SEC AI Governance Framework 2569, the OIC Insurance AI Guideline 2568, ETDA GenAI and draft EIA Playbook guidance, and NCSA guidelines alongside technical security testing recommendations.

Overview
Target Audience
Financial institutions, capital market operators, insurers, and payment providers referencing BOT, SEC, OIC, NCSA, and ETDA guidance
Testing and Review Frequency
Select testing and review cycles for risk, AI use, and data or model changes, distinguishing guidance from binding obligations: BOT Wor 5994/2568, the SEC 2569 principle-based framework (Clause 2.4.7(3)), and OIC 2568 Soft Regulation guidance (Clause 2.2.2) recommend risk-proportionate testing and adversarial evaluation rather than imposing uniform statutory cycles.
Assessment Scope
Data, model, and cyber risks, SEC example AIMS controls and impact assessments, OIC Soft Regulation AI robustness and vulnerability testing guidance, NCSA AI lifecycle security, and ETDA draft EIA playbooks
Relevant Risks
Data leakage, erroneous outputs, model bias and drift, unauthorized AI tool execution, and governance gaps where human oversight is absent

BOT policy guideline Wor 5994/2568 and its scope

Document 25680178 is a policy guideline for financial service providers to reference and apply according to their AI use. It is not a continuous penetration-testing mandate for all organizations.

Thai AI governance: 5 authorities to test evidence: BOT, SEC, and OIC, ETDA guidance, NCSA guidance, Assess by risk
Thai AI governance: 5 authorities to test evidence

Check sector applicability across BOT, SEC, OIC, ETDA, and NCSA guidance vs binding rules before setting test scope.

  • BOT, SEC, and OIC — Wor 5994/2568, Capital Markets 2569, and Insurance 2568
  • ETDA guidance — GenAI Governance and the draft EIA Playbook
  • NCSA guidance — Security controls across the AI lifecycle B.E. 2568
  • Assess by risk — AI pentesting, red teaming, and remediation verification
  • The scope includes financial institutions, specialized financial institutions, and regulated payment system and payment service operators.
  • The guideline identifies data risks, model-development risks and cyber threats, with management proportionate to the AI use case.
  • Consider data quality and security, model testing, monitoring and third-party risks alongside the other applicable rules.
  • Select technical tests and review cycles for actual risk and applicable obligations. This policy guideline does not itself establish a fixed testing cycle or a new penalty regime.

SEC AI Governance Framework for Capital Markets (August 2569)

The SEC Thailand 1st Revised Edition (August 2569) provides a hybrid principle-based and risk-based framework for capital market operators rather than a one-size-fits-all mandate.

SEC capital markets and OIC insurance AI governance frameworks: SEC Five Core Elements, SEC Clause 2.4.7(3) and Annexes, OIC Soft Regulation (4 Parts), OIC Clauses 2.2.1, 2.2.2 and 2.2.7
SEC capital markets and OIC insurance AI governance frameworks

Both the SEC August 2569 guidance and the OIC November 2568 Soft Regulation scale technical AI controls and adversarial testing to system impact rather than imposing one uniform checklist.

  • SEC Five Core Elements
    Covers Governance, Principles, Risk Management, AIMS Control, and Enablers
  • SEC Clause 2.4.7(3) and Annexes
    Recommends pre- and post-launch red teaming alongside 12-step AIA, Risk Rubric, and 8 AIMS domains
  • OIC Soft Regulation (4 Parts)
    Spans Governance, Robustness and Security, Transparency, and Fairness
  • OIC Clauses 2.2.1, 2.2.2 and 2.2.7
    Covers Secure by Design, risk-based robustness tests, GenAI safeguards, and Part 4 fairness checks
  • Encompasses 5 core elements: Governance, Principles, Risk Management, AIMS Controls, and Enablers, scaling controls to AI use-case risk.
  • Clause 2.4.7(3) recommends Red Team and Adversarial Testing both before and after deploying an AI system.
  • Annex 2 outlines a 12-step AI Impact Assessment flow, while Annex 4 provides an AI Inventory template and Risk Rubrics classification.
  • Annex 5 provides example role-based AIMS Controls across 8 categories for Developer, User, and Customizer roles (referencing NCSA AI Security guidance for cybersecurity depth), and Annex 13 provides example disclosure and anti-AI-washing guidance.

OIC AI Governance Guidelines for Insurance Companies (November 2568)

The OIC AI Governance Guidelines (November 2568) adopt a Soft Regulation and risk-based proportionality approach rather than a binding notification for insurers.

  • Covers 4 core parts: AI Governance, Robustness & Security, Transparency, and Fairness.
  • Clause 2.2.1 recommends Secure by Design principles, and Clause 2.2.2 recommends AI security and vulnerability testing commensurate with system risk.
  • Clause 2.2.7 sets out risk-management practices for Generative AI adoption, including data leakage safeguards and output review.
  • Part 4 addresses bias management and policyholder protection regarding granular risk-based pricing.

ETDA GenAI governance and the draft EIA Playbook

The 2024 GenAI Governance Guideline supports organizational governance. The referenced EIA Playbook remains marked as a draft dated 11 September 2026.

  • Assess intended GenAI use, accountable roles, and data, output and operational risks before procurement and adoption.
  • EIA references ten UNESCO ethical principles. Scoping addresses two, and principles-based assessment addresses the remaining eight across seven categories before impact mapping.
  • Categories cover safety, fairness, sustainability, privacy, human oversight, accountability, and a combined category for transparency and explainability with awareness and literacy.
  • EIA evaluates socio-technical impacts and safeguards. It is not an AI penetration-test certificate, and a draft is not enacted law.

NCSA guidance for AI security across the lifecycle

Use NCSA guidance to support governance and security across the AI lifecycle. Check document scope and status before treating guidance as an organizational obligation.

  • Identify accountable owners, assets, data and AI risks, including external integrations.
  • Plan controls for data, models, deployed systems, monitoring and incident response for the organization context.
  • Connect testing evidence to remediation, risk review and system changes. See the linked NCSA AI Security guide for source-specific detail.

Recommended AI penetration testing and red teaming

STH technical assurance recommendations provide risk-based testing evidence to complement governance and remain distinct from BOT, SEC, OIC, ETDA, and NCSA guidance, none of which imposes a uniform mandatory checklist across every organization.

Risk-based technical AI testing evidence: LLM Prompt Injection and Tools, RAG Access and Tenant Isolation, Model Quality vs Security Findings, Safe Stop, HITL and Retest
Risk-based technical AI testing evidence

Accountable system owners align acceptance thresholds, evidence depth, safe-stop conditions, and retesting to real business and safety impact.

  • LLM Prompt Injection and Tools
    Test direct and indirect prompt injection, disclosure boundaries, and authorized tool limits
  • RAG Access and Tenant Isolation
    Verify source-document authorization, tenant isolation, and synthetic ingestion controls
  • Model Quality vs Security Findings
    Separate accuracy, bias, and drift checks from security vulnerabilities and access failures
  • Safe Stop, HITL and Retest
    Set acceptance criteria, human oversight thresholds, safe-stop triggers, and retesting
  • For LLM applications, test direct and indirect prompt injection, disclosure boundaries and authorized tool use.
  • For RAG, verify source-document authorization, tenant isolation and knowledge-base ingestion using synthetic data.
  • Test accuracy, bias and drift for relevant use cases. Separate model-quality results from vulnerability and access-control findings.
  • Define acceptance criteria, human oversight, safe stopping where needed and retesting. Accountable owners should decide using evidence and impact.

References and Assessment Scope

Summary of source references, assessment scope, and review timing in the context of each item.

Reference Topic Assessment Scope Testing Cycle
BOT policy guideline Wor 5994/2568 BOT policy guideline Wor 5994/2568 and its scope Document 25680178 is a policy guideline for financial service providers to reference and apply according to their AI use. It is not a continuous penetration-testing mandate for all organizations. According to applicable scope and service risk, not a universal testing cycle
SEC AI Governance Framework August 2569 (Principle-based Guideline) SEC AI Governance Framework for Capital Markets 5 core elements, Clause 2.4.7(3) Red Team and Adversarial Testing, Annex 2 (12-step AI Impact Assessment), Annex 4 (AI Inventory), and Annex 5 (example AIMS Controls across 8 categories) Risk-proportionate principle-based cadence (for example before and after deployment or on material system change, as defined by the operator)
OIC AI Governance Guidelines November 2568 (Soft Regulation) OIC AI Governance Guidelines for Insurance Companies 4 core parts, Clause 2.2.1 Secure by Design guidance, Clause 2.2.2 risk-based AI security and vulnerability testing guidance, Clause 2.2.7 GenAI controls, and bias management Recommended testing and review proportionate to system risk under Soft Regulation guidance (not a statutory annual mandate)
ETDA GenAI 2024 and draft EIA Playbook ETDA GenAI governance and the draft EIA Playbook The 2024 GenAI Governance Guideline supports organizational governance. The referenced EIA Playbook remains marked as a draft dated 11 September 2026. According to applicable scope and service risk, not a universal testing cycle
NCSA AI Security Guidelines 2568 NCSA guidance for AI security across the lifecycle Use NCSA guidance to support governance and security across the AI lifecycle. Check document scope and status before treating guidance as an organizational obligation. According to applicable scope and service risk, not a universal testing cycle
STH assessment recommendations Recommended AI penetration testing and red teaming STH technical assurance recommendations provide risk-based testing evidence to complement governance and remain distinct from BOT, SEC, OIC, ETDA, and NCSA guidance, none of which imposes a uniform mandatory checklist across every organization. According to applicable scope and service risk, not a universal testing cycle

Implementation Checklist

Select items your organization has completed. The percentage shows selected items only and does not assess security or compliance.

0%

Frequently Asked Questions (FAQ)

Key answers and practical guidance addressing common compliance questions.

Who is within BOT Wor 5994/2568 scope?

The scope includes financial institutions, specialized financial institutions, and regulated payment system and payment service operators. Document 25680178 is a policy guideline for financial service providers to reference and apply according to their AI use. It is not a continuous penetration-testing mandate for all organizations.

Is the EIA Playbook law or an AI penetration-test certificate?

EIA evaluates socio-technical impacts and safeguards. It is not an AI penetration-test certificate, and a draft is not enacted law. EIA references ten UNESCO ethical principles. Scoping addresses two, and principles-based assessment addresses the remaining eight across seven categories before impact mapping.

How should third-party AI integrations be assessed?

For LLM applications, test direct and indirect prompt injection, disclosure boundaries and authorized tool use. For RAG, verify source-document authorization, tenant isolation and knowledge-base ingestion using synthetic data.

What technical evaluation guidance does the SEC AI Governance Framework (August 2569) provide?

The SEC Thailand 1st Revised Edition (August 2569) is a hybrid principle-based and risk-based framework rather than a one-size-fits-all mandate. Clause 2.4.7(3) recommends Red Team and Adversarial Testing before and after AI deployment, Annex 2 outlines a 12-step AI Impact Assessment, Annex 4 provides an AI Inventory template, Annex 5 provides example role-based AIMS Controls across 8 categories (referencing NCSA AI Security guidance), and Annex 13 provides anti-AI-washing disclosure examples.

How does the OIC AI Governance Guideline (November 2568) approach AI security testing?

The OIC guideline (November 2568) is Soft Regulation rather than a binding notification. Clause 2.2.1 recommends Secure by Design principles, and Clause 2.2.2 recommends conducting AI security and vulnerability testing commensurate with system risk, alongside GenAI safeguards (Clause 2.2.7) and Part 4 model bias controls for granular risk-based pricing.

Official source documents

BOT Wor 5994/2568 policy guideline and AI governance references

Go to the official source

Request a quote