Vietnam's AI Law Takes Effect: What Legal Teams Need to Know Now
Vietnam's artificial intelligence law is now in force as of March 1, making it the first comprehensive AI framework in Southeast Asia. The statute zeroes in on risks from generative AI, mandates human oversight, and mirrors several themes seen in the European Union's approach.
The law reaches developers, providers, and deployers operating in Vietnam, including foreign entities with in-country operations. Its immediate signal: AI products and customer interactions must be truthful, traceable, and under human control.
Scope and reach
The statute applies across the AI supply chain-builders, distributors, and users deploying systems in production. It expressly covers Vietnamese organizations and foreign companies doing business in Vietnam.
Expect sectoral guidance and implementing decrees to refine details. Until then, assume core obligations apply to both B2C and B2B settings where content is generated or customers interact with automated agents.
Core obligations highlighted by the law
- Human oversight and control: Maintain a clear ability to supervise, intervene, and override AI outputs-especially for generative systems.
- Labeling of AI content: Clearly mark AI-generated media, including deepfakes or content that could be mistaken for real people or events.
- AI interaction disclosure: Tell users when they are dealing with an artificial agent rather than a human.
- Risk focus on generative AI: Address risks such as misinformation, online abuse, and copyright concerns through documented controls.
Alignment with global regimes
Vietnam's framework echoes the EU's risk-based model that stresses transparency and human control. If you already track EU developments, you'll recognize familiar concepts and can reuse parts of your governance stack.
For context, see the European Commission's overview of the EU AI Act. South Korea's AI law reportedly took full effect in January, and the EU's rules will phase in through 2027-Vietnam is moving in step with this global push.
Government build-out
The government plans a national AI computing center, stronger data resources, and Vietnamese large language models. This sits within an economic strategy targeting double-digit growth, with AI and data positioned as key pillars.
Enforcement reality
Analysts describe the statute as a strong starting point. The actual bite will come from implementing decrees, sector-specific rules, and day-to-day enforcement.
Until guidance lands, companies will face some uncertainty around scope, thresholds, and technical standards. That's not a reason to wait-set pragmatic controls now and adjust as details firm up.
Action checklist for legal and compliance (next 90 days)
- Map systems and roles: Inventory AI use cases that generate content or touch customers. Identify your role per use case (developer, provider, deployer).
- Gap assess core duties: Review human oversight, labeling, and disclosure across products, marketing, support, and trust & safety flows.
- Labeling/disclosure policy: Draft and publish a policy covering deepfake labeling and agent disclosure. Localize notices in Vietnamese where needed.
- Content provenance: Implement practical signals (e.g., watermarks or content metadata) so AI outputs are consistently identifiable across channels.
- Process controls: Require human review for high-impact or sensitive outputs (news-like content, political content, safety-critical instructions).
- Contract updates: Add supplier obligations for labeling, data rights, training-data assurances, and timely cooperation on takedowns or corrections.
- Records and audits: Document risk assessments, oversight procedures, and change logs for models and prompts used in production.
- Frontline training: Equip support, sales, and content teams with scripts that clearly disclose AI use and escalation paths to a human.
- Monitor guidance: Track forthcoming decrees and sectoral rules; prepare to adapt templates and control libraries quickly.
Open questions to watch
- How "generative AI," "deepfake," and similar terms will be defined in guidance.
- Technical standards for labeling and provenance, and whether specific formats will be mandated.
- Expectations on dataset sourcing and copyright, including notice and takedown procedures.
- Any registration, notification, or pre-market review processes that may be introduced by decree.
- Penalty ranges, supervision model, and enforcement priorities by sector.
Why this matters for foreign companies
If you operate in Vietnam, these duties attach to customer touchpoints, product UX, and distribution channels. Plan for Vietnamese-language disclosures, consistent labeling across media, and a clear path to human support.
Bottom line
The law sets the destination: human-controlled, clearly labeled, accountable AI. While the fine print is still coming, the practical work-mapping systems, setting controls, and updating contracts-should start now.
Further resources
Your membership also unlocks:
