Decree 33 names 46 AI systems across six sectors, with transportation accounting for 31, turning Vietnam's framework law into enforceable compliance obligations.
Vietnam's first high-risk AI catalog is dominated by autonomous transport systems, signaling where Hanoi is drawing its regulatory red lines as the country's AI law moves from framework to enforcement.
The government released Decree 33 on 30 June 2026, naming 46 AI systems classified as high-risk under the 2024 Law on Artificial Intelligence. The list, signed by Deputy Prime Minister Ho Quoc Dung, spreads across six domains: education, ethnicity and religion, healthcare, banking, litigation, and transportation.
Transportation alone accounts for 31 of the 46 entries. The list includes automated vehicle control systems operating without human oversight and automated traffic and signal-directing systems, according to Vietnam News (VNA) state press coverage. The other five sectors share the remaining 15 systems. The category mix, weighted heavily toward physical mobility, positions Decree 33 as a critical-infrastructure-and-public-trust regulation rather than a general-purpose AI bill.
That calibration reflects a coherent theory of harm. Vietnam's Law on Artificial Intelligence defines high-risk AI as systems that may cause significant harm to lives, health, legitimate rights or interests of organizations or individuals, or to national interests, public interests, and national security. Automated traffic and autonomous-driving systems sit closest to that definition's core: a malfunctioning model can kill people directly, and the consequences are visible, physical, and attributable. Generative systems that produce biased text or deepfakes rarely carry the same immediate kinetic risk.
Vietnam is not alone in choosing this calibration, but its prioritization differs from the European Union's AI Act and from the United States' sector-by-sector approach. The EU's high-risk list is broader and includes biometrics, recruitment, and critical infrastructure management. US regulation has leaned on existing sector regulators, from NHTSA for transportation to the FDA for healthcare, rather than naming specific AI systems. Vietnam's decree sits between the two, issuing a specific catalog of named systems but skewing toward physical automation rather than generative content.
Providers and operators must report the risk level of their AI systems to the Ministry of Science and Technology before deployment, complete a conformity assessment before launch, maintain that assessment throughout the lifecycle, and keep an active risk management plan during operation. The decree takes effect on 15 August 2026, with transitional periods running until 1 September 2027 for education, healthcare, and banking systems that have gone into operation before the decree. The framework explicitly covers six domains, with transportation accounting for the largest share.
For companies building or deploying the named systems, the open question is how the conformity assessment will work in practice. Decree 33 names the categories but not the assessment bodies, the audit cadence, or the timelines. Independent industry or academic reaction to the decree has not yet surfaced in English-language coverage, leaving the practical burden on operators uncertain until further guidance is issued.
The Ministry of Science and Technology's implementing guidance is the next milestone to watch; until it lands, the list sets the categories and the regulatory weight behind them, without yet setting the pace.