You are using an outdated browser and your browsing experience will not be optimal. Please update to the latest version of Microsoft Edge, Google Chrome or Mozilla Firefox. Install Microsoft Edge

June 19, 2025

Thailand Drafts AI Risk Management Guidelines for Financial Service Providers

Subscribe to Updates

The Bank of Thailand (BOT) has released draft guidelines establishing principles for managing artificial intelligence (AI) risks in the financial sector. The draft guidelines provide a structured framework for the responsible adoption of AI technologies. Financial service providers will be able to use the guidelines as a reference to appropriately manage their risks in a manner that aligns with internationally recognized best practices.

The BOT is accepting public comments on the draft guidelines until June 30, 2025.

Scope and Application

The draft guidelines apply to all financial service providers, including financial institutions and special financial institutions under the Financial Institution Business Act, as well as payment providers under the Payment Systems Act. These guidelines supplement existing BOT risk management guidelines covering IT risk management, third-party risk management, data governance, and market conduct.

The guidelines define AI systems as systems that mimic human intelligence, including machine learning, deep learning, generative AI (such as large language models), and agentic AI. This definition specifically excludes rule-based automation systems like robotic process automation and condition matching.

Key Risk Management Principles

The guidelines lay out two main principles in managing AI risk.

  1. Governance: Financial service providers should define and establish clear roles and responsibilities for their personnel and AI system supervision structures to uphold FEAT (fairness, ethics, accountability, and transparency) principles as follows:
    • Stakeholder roles and responsibilities. Financial service providers should define roles and responsibilities for boards and executives on AI risk oversight. Responsibilities include establishing an AI system usage policy, designating personnel responsible for AI risk management, and building awareness of AI-related risk within the organization.
    • AI system usage policy. The AI system usage policy should be aligned with organizational objectives, regulatory requirements, and FEAT principles. These policies should be reviewed regularly to respond to technological advancements and evolving risk profiles.
    • Risk management throughout the AI lifecycle. Risk management should encompass the entire AI lifecycle, from establishing risk appetite to implementing continuous risk assessment and control measures tailored to specific use cases. When AI systems are used in strategic functions or customer interactions (e.g., loan approval, account opening), human oversight must be integrated into decision-making processes. In customer interactions with AI systems, customers should be notified and have options to disable or bypass AI features.
  1. Development and security controls: Financial service providers should have risk controls covering the AI development and deployment lifecycle as follows:
    • Data risk. Financial service providers should have measures to assess and ensure the quality, accuracy, currency, volume, and diversity of data used in AI model training. They should also implement data leakage prevention measures.
    • Model development risk. Financial service providers should have (1) clear evaluation metrics for assessing model accuracy and reliability through ongoing testing and monitoring both before and after deployment and (2) measures to ensure the explainability of AI outcomes. For generative AI applications, there should be specific measures to reduce AI hallucination risks.
    • Cybersecurity risk. Financial service providers should have measures to prevent and detect emerging cyber threats targeting AI systems, based on established standards such as the OWASP Machine Learning Security Top 10.

For more details on any aspect of fintech, technology, and cybersecurity in Thailand, please contact Athistha Chitranukroh at [email protected], Nopparat Lalitkomon at [email protected], Pornpan Wichawut at [email protected], Napassorn Lertussavavivat at [email protected], or Rujaporn Paritsantik at [email protected].

Related Professionals

Industries

Fintech

Technology

Practices

Banking and Finance

Location

RELATED INSIGHTS​ 

June 23, 2026
On May 26, 2026, Thailand’s Department of Land Transport (DLT) published for public consultation a draft amendment to the Ministerial Regulation on Electronic Ride-Hailing Vehicles that would, for the first time, allow juristic persons (legal entities) to register vehicles as electronic ride-hailing cars—a right that currently belongs exclusively to natural persons, limited to one person per one vehicle. If finalized in its current form, the regulation would significantly expand the supply side of Thailand’s ride-hailing market by enabling corporate fleet operators to enter the space. The public comment period is open through June 24, 2026. Key Principles Under the Draft Regulation Under the proposed amendment, juristic persons that maintain a fleet of at least 50 vehicles will be permitted to register vehicles as electronic ride-hailing cars. This represents a fundamental shift from the current framework, which restricts registration to individual natural persons on a one-person-one-car basis. Vehicle Specifications Corporate-owned ride-hailing vehicles must meet the following requirements: Be brand new from the factory, or no more than two years old from first registration with no more than 20,000 km of use. Not be a vehicle that has been reconstructed or repaired after involvement in a serious accident affecting safety—a standard consistent with public transport vehicles (RorYor. 6). Be classified as small, medium, or large in accordance with ministerial or director-general specifications. The vehicles may be equipped with safety devices such as interior or exterior cameras (video/photo recording) and can retain the original factory color of the vehicle body (no mandatory color change is required). License Plates Corporate ride-hailing vehicles will use license plates of the same size, characteristics, and color as those for private passenger vehicles not exceeding seven seats (RorYor. 1), rather than public transport plates. Potential Impact The government has stated that the regulation is intended to: Promote
Read More
June 23, 2026
On May 14, 2026, Thailand published a ministerial regulation in the Government Gazette to prescribe measures for prevention and suppression of technology crimes. The regulation creates a comprehensive procedural framework for returning money and digital assets to victims of technology crimes. It will take effect 90 days after publication (in mid-August 2026), giving affected entities a limited window to prepare. Mandatory Reporting Obligations for Financial Institutions When a deposit account, e-money account, or digital asset wallet is frozen in connection with a technology crime, the relevant financial institution or business operator must report transaction data to the Anti-Money Laundering Office (AMLO) via AMLO’s designated electronic system. Required data elements include account numbers (sender and receiver), names, identification or passport numbers, legal entity registration numbers, phone numbers, remaining balance, damage amount, transaction reference numbers, and the bank case ID. Institutions that already share data through the information-sharing system under the emergency decree are deemed to have satisfied this reporting obligation, creating an incentive for platform participation. When the Royal Thai Police or the Department of Special Investigation seize or freeze assets related to technology crimes, they must provide AMLO with investigation reports, complaint evidence, money-trail data, and account statements. Notification and Claims Process Once the AMLO secretary-general approves verified reports of a technology crime, the account information of persons connected to the crime will be published in the Government Gazette, triggering a 90-day window for victims to file claims and for related persons to file objections. Officers will also publish details on AMLO’s electronic media and send registered mail to identified victims, which will be deemed received after 7 days domestically or 15 days internationally. Victims have 90 days from the date the crime is published in the Government Gazette to file claims through AMLO’s electronic system. Claims must include
Read More
June 15, 2026
The surge in AI development has led to a desperate demand for large, high-quality training data. However, real-world data can be expensive to collect, difficult to access, and often subject to strict privacy and regulatory constraints. Synthetic data, which consists of artificially generated records that replicate the statistical properties of real-world data without reproducing specific individuals’ information, provides an appealing solution by generating artificial datasets at scale without relying on identifiable personal information. It combines speed, cost efficiency, and regulatory compliance, making it a sensible alternative for organizations seeking to reduce risks while maintaining data utility. When properly anonymized, synthetic datasets may fall outside the scope of laws such as the EU’s General Data Protection Regulation (GDPR) or Thailand’s Personal Data Protection Act (PDPA), reducing compliance burdens while still supporting high-quality model training. However, relying on synthetic data without rigorous legal due diligence could be a strategic mistake. It replaces one set of known risks (scraping, direct privacy liability) with a new set of complex liabilities. The narrative that synthetic data is a “silver bullet” for privacy and IP compliance is dangerous and could be misleading. While synthetic data addresses data scarcity, it also introduces new legal uncertainties. Legal counsel should anticipate downstream risks arising from compromised data sources. Models trained on unlawfully obtained data may need to be decommissioned, even if their outputs appear lawful. What is synthetic data? Synthetic data refers to artificially generated information created using AI techniques such as deep learning and generative models. Instead of copying real records, it reproduces the statistical patterns and relationships found in the original dataset. Synthetic data generally falls into three categories: Fully synthetic data – Entirely new data points generated from learned patterns. The model studies the structure of the original data and produces records that resemble real-world
Read More
June 11, 2026
Thailand’s Electronic Transactions Development Agency (ETDA) has released a revised draft Electronic Transactions Act (ETA) for public hearing from May 12, 2026, to June 15, 2026. This is not merely an amendment to certain provisions of the current ETA, but a comprehensive redrafting of the entire act. The revised draft ETA introduces several significant changes from the current framework, with practical implications for businesses operating in Thailand. Unified Coverage of Public and Private Sectors The current law segregates government transactions into a separate chapter with distinct rules. The draft ETA eliminates this division, defining “transaction” to encompass civil and commercial juristic acts as well as administrative procedures, administrative contracts, and other acts of government agencies. Enhanced E-Signature Definition The definition of “electronic signature” is broadened to expressly include biometric data and refocused on identifying the signatory and demonstrating intent regarding the content of the electronic data. Shift in Burden of Proof When a party challenges the reliability of electronic data created using a “trusted electronic method” or a method prescribed by the ETDA, the burden of proof and the cost of proving unreliability shifts to the challenger. Introduction of New Digital Method Concepts The draft ETA introduces several new digital method concepts that are not currently recognized under the existing ETA framework. These include: Electronic timestamping (e-timestamp) Electronic registered delivery Electronic company seals Electronic stamp duty compliance Electronic identity authentication and verification Electronic transferable records (electronic bills of lading, promissory notes, and similar negotiable instruments) Recognition of Automated Systems and Electronic Contracting The draft ETA expressly recognizes the legal validity and enforceability of contracts formed through automated systems, including contracts concluded entirely between automated systems or between an automated system and a person. A party may not deny the binding effect of such contracts solely because no human review
Read More