What did Vietnam’s Decree 142 require?

Decree 142: risk tiers, deepfake labeling, and chatbot disclosure

Vietnam has introduced Decree 142 to implement the country’s AI law, adding compliance obligations aimed at managing risks from AI systems used in the real world.

Companies that deploy AI in Vietnam now must classify AI models by risk level. The goal is to determine which safeguards apply based on how potentially harmful or sensitive the model’s outputs and behavior could be.

Key obligations spelled out

The requirements reported include:

• Classifying AI models by risk level
• Labeling deepfakes so people can tell when synthetic media is being presented
• Disclosing chatbot use, making it clearer when users are interacting with an automated system

Why it matters

These measures directly address two fast-growing issues in AI deployment: synthetic media used for manipulation, and conversational agents that can blur the line between human and machine interaction.

Labeling deepfakes is particularly important because deepfake content can be indistinguishable at a glance, making transparency a first line of defense for consumers and regulators.

Requiring disclosure of chatbot use also impacts safety and accountability. If users are told they are conversing with a bot, organizations can reduce deception risk and make complaint-handling and remediation pathways easier to establish.

Finally, risk-based model classification is a common regulatory approach intended to avoid one-size-fits-all rules. Instead of treating every AI system identically, the framework can scale obligations upward or downward depending on expected impact.

No further implementation details were provided in the summary beyond these core compliance points.