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

July 2, 2026

Thailand Releases New Draft Artificial Intelligence Act

Subscribe to Updates

Thailand’s Electronic Transactions Development Agency (ETDA) released a new version of the draft Act on Artificial Intelligence on July 2, 2026, for a public hearing period expected to be approximately 30 days. The draft act adopts a risk-based regulatory approach modeled in part on international frameworks—particularly the EU’s AI Act—while incorporating provisions tailored to Thailand’s regulatory landscape and digital economy objectives. If enacted in its current form, the law would introduce extraterritorial obligations, a tiered risk classification system, strict liability for AI-related damages, and new transparency requirements for AI-generated content.

Scope and Extraterritorial Application

The draft act applies to AI development, deployment, or any other action affecting people in Thailand, even if the action occurs outside the country. Of note:

  • This extraterritorial reach creates compliance obligations for global AI companies whose systems impact Thai residents or consumers, even if the provider has no physical presence in Thailand.
  • Foreign AI providers serving Thai deployers or users must appoint a local coordinator or authorized representative. Depending on the type of AI system, the representative may need full authority to act on behalf of the provider without any limitation of liability.
  • Certain activities are exempt from the draft act’s oversight, including AI used by natural persons solely for personal or household activities, AI for educational research conducted by higher education institutions with ethics committee approval, research and development activities conducted prior to distribution or service provision, and other AI systems prescribed by royal decree.

Risk-Based Classification Framework

The draft act establishes a tiered risk classification system with three main categories:

  • Prohibited AI. The act outright prohibits AI systems employing cognitive-behavioral manipulation using subliminal techniques, AI systems causing unfair broad-scale discrimination from processing irrelevant data, and other categories of serious risk as determined by announcement of a forthcoming committee that will be responsible for national AI planning.
  • High-risk AI. High-risk AI systems are those designated by royal decree as affecting national security, health, environment, energy, telecommunications, transport, or public utilities.
  • Designated AI systems. A subsequent royal decree may additionally require regulator notification, registration, or licensing of certain AI systems before deployment.

Obligations for High-Risk AI System Providers and Deployers

Providers of high-risk AI must develop systems that are efficient and fit for purpose, transparent in operation, subject to meaningful human control, fair and nondiscriminatory, and aligned with foreseeable risks. The regulator may announce risk oversight guidelines covering 13 areas, including risk management, bias prevention, cybersecurity, human oversight, transparency, and complaint handling.

Deployers must implement risk management systems, follow provider instructions, assign capable oversight personnel, mitigate damage from AI-related incidents, retain operational logs for a minimum period (six months under the Thai text), and notify authorities of unforeseen risks.

Transparency Obligations for AI-Generated Content

The draft act introduces transparency requirements for content generation by AI systems:

  • Developers of AI systems capable of generating or modifying images, audio, or video must assess risks, implement mitigation measures, and embed machine-readable marks to identify AI-generated content.
  • Persons who introduce AI-generated content relating to national security, election-related content, investment credibility, food or drug properties, impersonation, or illegal acts into public systems must disclose that it is AI-generated or AI-modified content.
  • Platform providers face specific obligations, including risk assessment, providing reporting channels, verifying AI content, displaying labels, and preparing annual operational summaries.

Data Localization and Contract-Controlled AI Businesses

The draft act grants the forthcoming national AI planning committee authority over data and contracting requirements in sensitive sectors:

  • The committee may designate “contract-controlled AI businesses” applicable to services provided to government or critical infrastructure agencies, and may prescribe mandatory contract terms addressing data processing, risk management, security, modification and termination procedures, and post-termination obligations.
  • The committee may also require data processing within Thailand for AI services of national importance. This introduces uncertainty for cloud-based AI services and may require infrastructure investment in Thailand.

Strict Liability Regime

The draft act imposes joint liability for damages regardless of willful act or negligence—a strict liability standard. Defenses are limited to force majeure, the victim’s own act or omission, or compliance with an official order.

Enforcement and Penalties

The draft act establishes an enforcement framework, with escalating consequences for noncompliance:

  • The regulator may order providers and deployers to rectify insufficient measures.
  • With ministerial approval, the regulator may petition the court to order temporary service suspension, product recall, or deployment suspension.
  • If providers fail to comply, the regulator may petition the court to order ISPs to block AI system dissemination in Thailand.

Administrative fines range from THB 1 million to THB 5 million, depending on the nature of the violation.

AI Product Launches, Regulatory Sandbox, and Self-Regulation

The draft act also establishes a regulatory sandbox framework for testing AI systems in regulated sectors, data-sharing infrastructure through Thailand’s Big Data Institute (a national public organization), and frameworks for self-regulation and best practices. Compliance with self-regulatory best practices may also serve as selection criteria for government investment promotion programs.

Implementation Approach and Next Steps

There will be a phased implementation of the draft act’s measures. Core measures related to the launching of AI products take effect immediately upon publication. Risk control, supervision, and serious incident provisions take effect 180 days after publication, providing a compliance preparation window for affected businesses.

Organizations that develop, deploy, or rely on AI systems affecting persons in Thailand should begin assessing their compliance. They are also encouraged to submit comments on the draft act to the Ministry of Digital Economy and Society during the public hearing period.

Related Professionals

Industries

Technology

Location

RELATED INSIGHTS​ 

July 27, 2026
A new decree on penalties for violations related to the crypto asset market creates compliance risks for offshore crypto asset exchanges in Vietnam that do not hold, and practically cannot obtain, a Vietnamese license, and for Vietnamese users who continue to transact on those platforms. Decree No. 284/2026/ND-CP (Decree 284), issued by the government of Vietnam on July 16, 2026, formally establishes an administrative penalty framework for violations related to crypto assets and the crypto asset market. The decree takes effect on September 1, 2026, and will remain in force for the duration of the five-year pilot program under Resolution No. 05/2025/NQ-CP, which is scheduled to end in September 2030. Direct Penalties on Vietnamese Users The most immediate commercial risk to offshore platforms is that their Vietnamese users now face direct personal liability for using their exchanges. Vietnamese users who trade crypto assets outside of a Ministry of Finance-licensed service provider face fines of up to VND 50 million (approximately USD 1,900). Vietnamese users trading in crypto assets that are offered or issued to foreign users face higher penalties of up to VND 100 million (approximately USD 3,800). It is expected that Vietnamese users will be more willing to migrate away from offshore platforms now that there is a risk of real enforcement against them. Penalties on Unlicensed Service Providers Violations of providing crypto asset services or advertising crypto-related services without a license face fines of up to VND 200 million (approximately USD 7,700). Operating a crypto asset trading market without proper authorization falls within the same highest penalty bands. Organizations that violate issuance, provision, or disclosure rules may face fines of up to VND 200 million. Although the maximum administrative fine per violation is capped at VND 200 million for organizations and VND 100 million for individuals, these
Read More
July 21, 2026
Thailand’s Ministry of Digital Economy and Society (MDES) published a notification establishing an expedited court-ordered takedown mechanism for online content in cases of “urgent necessity.” The notification, which was issued on July 17, 2026, under the Computer Crime Act B.E. 2550 (2007), as amended, took effect the following day. It significantly expands the categories of content subject to rapid government-initiated removal. Content Categories Subject to Takedown The notification defines “urgent necessity” (section 20, paragraph 5, of the Computer Crime Act) as circumstances where any delay in suppressing computer data may impact national security, religion, the monarchy, good morals, social culture, or public order. In this regard, it establishes four broad categories of content: Computer Crime Act offenses. National security offenses. IP and other criminal offenses, where it is contrary to public order or good morals and a competent officer has requested its suppression. Content contrary to public order or good morals, a broad residual category encompassing 14 subcategories approved by the Computer Data Screening Committee. The fourth category is the most expansive. Its 14 subcategories include: Content defaming, mocking, satirizing, or devaluing the monarchy. Online gambling advertising or facilitation. Offering illegal firearms for sale. Offering baraku (hookah) products or e-cigarettes for sale. Offering cannabis inflorescences or processed cannabis products for sale. Advertising or soliciting prostitution. Content inciting violence, hatred, or social division. Unauthorized overseas employment advertising. Offering boiled kratom juice for sale. Online sale or advertising of alcoholic beverages. Content satirizing or degrading Buddhism. Money lending at interest rates exceeding legally prescribed limits. Advertising or disseminating information about surrogacy services. Forgery of documents, cards, or official documents. Enforcement Procedure In cases of urgent necessity, a competent official assigned by the MDES permanent secretary must file a petition with supporting evidence to the court with jurisdiction, requesting an order to
Read More
July 20, 2026
On July 16, 2026, Thailand’s Personal Data Protection Committee (PDPC) published a notification in the Government Gazette establishing detailed rules governing data subjects’ right of access under section 30 of the Personal Data Protection Act B.E. 2562 (2019) (PDPA). The notification will take effect 60 days after publication—mid-September 2026—giving data controllers a limited window to bring their processes into compliance. Scope The notification covers requests to access or obtain copies of personal data and requests for disclosure of the source of data collected without consent. Data subjects may exercise their rights directly or through authorized representatives. Key Requirements Important requirements set by the notification include the following: Required request channels. Controllers must provide at least two request channels: direct submission at the business location and registered mail. Electronic channels are optional but, if offered, may also be used for fulfilling requests. Request contents. Requests must be in writing or in electronic form and include the data subject’s name, the preferred access method, details of the data requested, and the requester’s signature. Controllers may request additional identifying information as needed. Identity and authority verification. Controllers may require official identity documents for verification. Authorized representatives must provide authorization documents and identity documents for both the data subject and the representative. Alternative verification methods (e.g., digital authentication) are permitted if they do not unreasonably obstruct data subjects’ rights. Review and response timelines. Controllers must review requests within 15 days. If the request is incomplete, the controller must notify the requester and allow at least 15 days to correct deficiencies. If not corrected, the request may be treated as abandoned. Once verified, controllers must fulfill requests within 30 days, extendable by another 30 days for large-volume or complex requests with notice to the requester. Methods for providing access or copies. Controllers may fulfill
Read More
July 16, 2026
Thailand’s Office of the Personal Data Protection Committee (PDPC) published a series of draft guidance documents for public consultation on July 7, 2026. Issued under the Personal Data Protection Act B.E. 2562 (2019) (PDPA), the drafts address a range of compliance issues and offer insight into the regulator’s current enforcement priorities. This article examines two of those drafts: one on lawful bases for processing personal data, and another on marketing and direct marketing. Together, they reflect the Office of the PDPC’s evolving expectations on lawful-basis selection, accountability, and the use of personal data in marketing. Organizations operating in Thailand should assess the practical implications now, before the guidance is finalized. Lawful Bases: A Structured Selection Process The draft guidance on lawful bases introduces a systematic five-step process for selecting an appropriate lawful basis for each processing activity. Organizations are expected to: Identify the processing activity involved. Assess the appropriate lawful basis. Evaluate whether the data is necessary for the processing. Conduct a legitimate interest assessment (LIA) where applicable. Ensure transparency through privacy notices. The guidance provides practical explanations and examples for each lawful basis under section 24 of the PDPA—including archiving, research, statistics, vital interests, contractual necessity, legal obligation, public task, legitimate interests, and consent—as well as the bases applicable to sensitive personal data under section 26. The aim is to promote more consistent and accurate lawful-basis selection across public- and private-sector organizations. A recurring theme throughout the guidance is that organizations should select the lawful basis that most accurately reflects the actual purpose and circumstances of the processing activity. The guidance cautions against treating consent as a default or catch-all basis where another lawful basis is more appropriate. For processing based on legitimate interests, organizations should conduct and document an LIA. Processing involving sensitive personal data may require
Read More