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 6, 2026

Vietnam Introduces Sectoral List of High-Risk AI Systems

Subscribe to Updates

Vietnam has introduced an official list of high-risk AI systems, triggering more stringent compliance obligations for developers, suppliers, and deployers operating in the country. On June 30, 2026, the prime minister issued Decision No. 33/2026/QD-TTg (Decision 33), which establishes the List of High-Risk AI Systems under the Law on Artificial Intelligence (AI Law) and Decree No. 142/2026/ND-CP (Decree 142). Decision 33 takes effect on August 15, 2026.

Decision 33 is significant because only AI systems included on the list will be subject to the heightened compliance obligations applicable to high-risk AI systems under the AI Law and Decree 142. These include, among others, local presence requirements for foreign providers, mandatory conformity assessment before deployment, comprehensive risk management and data quality documentation, and strict liability for damages even when the provider is fully compliant. Decision 33 also specifies the applicable conformity assessment pathway for each listed system, indicating whether the system must undergo mandatory third-party conformity certification before being placed into use, or whether the provider may self-assess conformity or voluntarily engage a registered or recognized conformity assessment body.

Which AI Systems Are Covered?

Decision 33 identifies high-risk AI systems across six sectors—the key attributes of which are summarized below.

  • Education: AI systems used for automated assessment, learner ranking, behavioral monitoring, or generating educational content from uncontrolled data sources.
  • Ethnic affairs and religion: AI systems used to automatically score, classify, or rank applications for government ethnic policies; approve or reject regulatory applications; suspend benefits on suspicion of fraud; allocate budgets; or infer and classify individuals by ethnicity or religion for administrative purposes.
  • Healthcare: AI-assisted surgical systems and autonomous AI-powered surgical robots.
  • Banking: AI systems that autonomously conduct electronic banking transactions or make credit approval decisions.
  • Judicial proceedings: Certain large-scale biometric identification systems used in public-interest civil proceedings.
  • Transport: Thirty-one categories of AI systems, including highly autonomous driving systems, AI-based traffic signaling and dispatch, and AI systems controlling critical transport infrastructure.

Many listed systems fall within the high-risk category only where AI outputs are used as the sole basis for decisions without meaningful human review, reflecting Vietnam’s risk-based approach to AI regulation.

Compliance Timeline

In general, AI systems—both those existing before Decision 33’s effective date and those deployed in the six months after the decision takes effect—must meet compliance obligations by March 1, 2027. However, AI systems in healthcare, education, and finance that are already in operation before August 15, 2026, are granted an additional transitional period, with a compliance deadline of September 1, 2027.

Outlook

Organizations developing, supplying, or deploying AI systems in Vietnam should review their existing and planned systems to determine whether they fall within the List of High-Risk AI Systems under Decision 33. Operators of listed AI systems should assess the resulting compliance obligations under the AI Law and Decree 142. Organizations should also use the transitional periods provided under Decision 33 to develop and implement compliance plans where necessary.

Related Professionals

Industries

Technology

Location

RELATED INSIGHTS​ 

August 11, 2026
On July 27, 2026, the State Bank of Vietnam (SBV) released a draft decree proposing amendments to Decree No. 52/2024/ND-CP dated May 15, 2024, on non-cash payments (Decree 52). The draft decree would amend 17 of Decree 52’s 38 articles, with several key changes directly affecting providers of intermediary payment service (IPS). The key proposed changes affecting IPS providers are outlined below. Streamlining IPS Licensing Procedures A central objective of the draft decree is to simplify regulatory procedures for IPS providers. Notably, it would significantly reduce IPS licensing documentation requirements by removing the need to submit enterprise registration certificates, investment registration certificates, and documents evidencing the qualifications of the legal representative and general director. Instead, the SBV would retrieve this information directly from national business registration and other specialized databases, requesting additional documents only where the relevant information cannot be verified electronically or is incomplete. The draft decree also removes the current limit of two rounds for dossier supplementation and shortens processing timelines for several IPS licensing procedures such as issuance, amendment, and reissuance of IPS licenses. The processing time for new IPS license applications would be thereby reduced from 90 to 60 working days. In addition, several continuing IPS business conditions would be removed. For example, IPS providers would no longer be required to maintain certain representations relating to corporate restructuring or the legality of contributed capital. Likewise, the IPS project plan (đề án) would become a one-time application document rather than an ongoing licensing condition. If retained in the final decree, this change could provide IPS providers with significantly greater flexibility to implement post-licensing technology upgrades, system integrations, and corporate restructuring transactions without needing to revisit the originally approved project plan. The draft decree also removes the requirement for the SBV to consult the Ministry of Public
Read More
August 10, 2026
Thailand has finalized its social media KYC (“know your customer”) rules under Notification of the Electronic Transactions Commission on Measures to Prevent Technological Crimes for Social Media Service Providers (No. 2), which was published in the Government Gazette on May 5, 2026, and will take effect on November 1, 2026. While an early draft of the notification proposed requiring social media platforms to arrange identification of every user account, the final notification is significantly more targeted, focusing on paid online advertising and advertiser identity verification. Though the regulatory initiative primarily aims to combat online fraud and technology-related crimes, it also has important consequences for intellectual property enforcement, because the verified platform records that will be generated under the new requirements can help IP rights holders to identify anonymous online infringers. Key Regulatory Mandates The notification requires social media service providers to verify the identity of advertisers before their paid advertisements are published and disseminated in Thailand through social media, regardless of whether the advertising fees come from the advertisers or third parties. Verification of an advertiser is valid for one year, after which verification would have to be performed again before the platform could publish additional paid advertisements from the advertiser. Permitted verification methods are specified under the notification. A platform may verify an advertiser by checking identity evidence and confirming the connection between the advertiser and that identity evidence, with the notification giving facial comparison against certain government-issued identity documents as an example. Alternatively, platforms may verify advertisers through a digital identity verification and authentication system with an identity-proofing assurance level not lower than the level prescribed by Thailand’s Electronic Transactions Commission. The notification further requires platforms to retain only the advertiser’s information necessary to identify the advertiser, beginning from the start of the advertising activity and for
Read More
August 10, 2026
On July 31, 2026, Thailand’s Big Data Institute (BDI) launched a public consultation on the principles of a proposed new data-sharing law, with comments accepted until August 31, 2026. If enacted, the law would establish Thailand’s first comprehensive framework for government and private-sector data sharing, creating a systematic, secure, and transparent regime to support analytics, policymaking, research, and innovation. Central Data-Sharing Platform The draft law establishes a central system for data sharing, managed by the BDI. Government agencies would be required to connect to the BDI’s Data Integration and Intelligence Platform (also referred to as D2), in accordance with the BDI’s rules and procedures. Five Dimensions of Data Sharing The draft law covers five key types of data sharing between government (G), businesses (B), and consumers (C): G2B: Private organizations may request government data specifically for research and development purposes. The BDI will assess the applicant’s data governance, security, and privacy capabilities whether such measures meet prescribed standards before forwarding the request to the relevant government agency within 90 days. Any dispute may be escalated to a newly established Data-Sharing Promotion Committee for final determination. G2G: Government agencies may request data from other agencies through the central system. The data-holding agency must respond within 90 days, taking legality, necessity, proportionality, public interest, and personal data protection into account. Disputes may be referred to the Data-Sharing Promotion Committee for adjudication. B2G: In emergency situations involving public safety, economic security, or disaster response, the Minister of Digital Economy and Society may require private entities to provide data through the central data-sharing system. Government agencies must specify the data requested, demonstrate its necessity and expected benefits, and request only data reasonably available to the data holder. Requests for personal data must be limited to the minimum amount necessary. B2C: Royal decrees may
Read More
August 10, 2026
Thailand’s Office of the Personal Data Protection Committee (PDPC) recently released draft guidance on records of processing activities (ROPA) for personal data controllers and processors under the Personal Data Protection Act B.E. 2562 (2019) (PDPA). The draft guidance, which was presented to the public on July 7, 2026, addresses both controller records of collection, use, and disclosure of personal data and processor records of processing activities carried out on behalf of controllers. If implemented, the guidance will significantly expand organizational expectations for ROPA preparation, maintenance, and use across all sectors. Key Takeaways The draft guidance contains several important implications for organizations subject to the PDPA: ROPA reframed as a core accountability tool. The guidance elevates ROPA from an administrative record to a central accountability mechanism, connecting controller duties with recordkeeping obligations. ROPA as a source for privacy notices and governance documents. ROPA should serve as the primary source for privacy notices and align with consent management, retention schedules, DPIAs, incident response plans, and vendor contracts. Expanded scope across all activities. ROPA must cover all processing activities across the organization—including security, finance, HR, and external contractors—with correct controller or processor classification for each. Ongoing maintenance and auditability. ROPA must be updated for any change to systems, purposes, or processors, reviewed at least annually, and maintained with version control and a designated owner. Enhanced vendor, processor, and cross-border transfer requirements. Organizations must document all processors, external recipients, and cross-border transfers, specifying purposes, access scope, and destination countries. Linkage with risk assessment, DPIAs, and LIAs. ROPA should assign risk levels to each activity and identify when data protection impact assessments (DPIAs) or legitimate interests assessments (LIAs) are required, functioning as a risk-management tool. ROPA and data breach readiness. Incomplete ROPA can delay breach response and notification. Organizations should map data flows, vendors,
Read More