Vietnam's AI Rules Are Here

Vietnam officially released Decree 142, the implementation decree for its AI regulations. Companies have to classify AI models by risk level, label AI-generated content, and let users know when a chatbot is at the other end of a convo.

The decree took effect on the 1st, but was officially released on the 7th of this month.

What's Required?

Risk Classification: There are three levels to this. AI systems must be classified as high, medium, or low risk. High-risk systems go on a Prime Minister's list (still in dev mode). Medium-risk systems require transparency and disclosure. Low-risk systems get light oversight.

Deepfake Labels: Anything simulating real people or recreating real-world events must be labeled clearly. Whether audio, images, videos, the labels must appear before or when users access the content, but not for basic edits like spelling or translations.

Chatbot Disclosure: Companies using AI chatbots must tell users they're talking to AI, not humans, especially when the convo could influence anything financial.

Incident Reporting: If AI causes a serious incident threatening cybersecurity or national security, the company must submit a preliminary report within 72 hours.

Foreign Companies: Global AI firms must appoint local reps to operate in Vietnam.

What's Banned?

Deceptive deepfakes used for impersonation or misinformation, including in contexts linked to geopolitical issues.

What Else?

A national AI portal and database to track AI systems operating in the country. The decree also includes incentives for domestic AI startups, like easier access to tech infrastructure and data.

The deepfake part is interesting. The decree has similarities with the EU AI Act, both have a “risk-based” approach without a one-size-fits-all model, but Vietnam's law bans specific actions (like harmful deepfakes) in a separate category instead of grouping them under an “unacceptable risk” category like the EU AI Act does.

Anyway. Go Vietnam!