Vietnam — AI Law Takes Effect: GenAI Transparency, Labelling & Oversight

⚠️

Action required. If you develop or deploy AI systems in Vietnam, prepare for product obligations around transparency, content labeling, and human oversight. Start with an inventory of AI features, user journeys, and vendor dependencies.

01 — Background

What is reported to be in force

Vietnam’s National Assembly adopted an AI law in late 2025, and reporting indicates it took effect at end-February 2026. The law targets the risks posed by generative AI. It also points to national AI infrastructure plans, including computing capacity and Vietnamese-language data and models. For organisations, the immediate compliance task is to operationalise transparency and control obligations in user-facing experiences.

Adoption

The law was passed in December (reported) and is positioned as a comprehensive AI framework.

Scope

Applies to developers, providers, and deployers, including foreign entities operating in Vietnam (reported).

Transparency

Users should be told when they are interacting with an AI agent rather than a human.

Labelling

AI-generated content, including deepfakes, must be clearly labelled when it cannot readily be distinguished from reality.

Oversight
Human oversight and control are stated themes, aligned conceptually with the EU AI Act approach.

Next
Implementing decrees and enforcement guidance will determine practical obligations and timelines.

🔍 Build an AI compliance baseline now

Baseline

Controls that are typically expected for transparency and high-risk generative use

🏷️

Output labeling

Consistent labels for AI-generated media and synthetic content across channels and export formats.

🧑‍⚖️

Human oversight

Escalation triggers, review queues, and release gates for higher-risk outputs or sensitive domains.

📊 Likely enforcement focus (qualitative)

Deepfake labeling & disclosureHigh

Visible, consistent user signals

User transparency for AI agentsMed-High

Chat, voice, support, and sales flows

Governance & oversight mechanismsMed-High

Roles, review triggers, logs

Documentation & evidenceMedium

Policies, testing, incident records

02 — Operating model

Operationalise transparency across devices and channels

GenAI features typically spread across web, mobile, APIs, and third-party integrations. Transparency fails when labels disappear during sharing, exports, or cross-platform reposts. Start with one policy and one implementation approach. Then test it in realistic user journeys and abuse scenarios.

🔑 Core principle: treat transparency as a product control. It must be visible, consistent, and hard to remove in normal usage. Where outputs can cause harm, add human oversight and response playbooks.

📱 Where the control must hold

📱

Mobile

Creation + sharing

→

💻

Web

Chat + export

→

🔌

API

Integrations

🏷️
Standardise labels. One format across UI, exports, and shared content.
🧪
Test with real flows. Share, repost, copy/paste, and screenshot scenarios.
🧑‍⚖️
Define oversight triggers. Impersonation, minors, and sexual content should trigger review or blocks.
🧾
Keep evidence. Logs for prompts, outputs, blocks, and escalations with access controls.
🧯
Run a response playbook. Reporting, takedown, and support steps for harmful synthetic media.

03 — Action plan

Four concrete actions to take now

Even before detailed decrees, you can make progress with a control baseline and an evidence pack. This reduces rework when implementation guidance lands and helps teams align across product, legal, and security.

ACTION 01
🗺️

Inventory AI features

Map AI use cases, user journeys, and vendors in Vietnam-facing products and services.

ACTION 02
🏷️

Implement labeling

Apply clear AI-generated content labels and ensure they persist across sharing and exports.

ACTION 03
🧑‍⚖️

Add oversight & safeguards

Define review triggers, block high-harm categories, and add escalation routes for edge cases.

ACTION 04
📋

Create an evidence pack

Document policies, testing, monitoring, and incident response readiness for regulatory questions.

⚠️ Three lessons for privacy teams

Lesson 1
Transparency is not a banner. It must survive cross-platform and cross-device distribution.

Lesson 2
Oversight needs criteria. Define triggers and owners before you need them.

Lesson 3
Implementation guidance will evolve. Build flexible governance and strong documentation now.

🛡️

Need a Vietnam AI compliance readiness sprint?

DPO Advisors can help implement a practical transparency and safeguards baseline for generative AI, and build evidence-ready governance for APAC rollouts.

Talk to our experts →

🛡️

DPO Advisors
Published February 2026 · dpoadvisors.com · contact@dpoadvisors.com