AI and copyright under the 2026 framework

·

AI is not a rights holder. Fully automatic output is unprotected. But the boundary of a human's "substantial creative contribution" is where the disputes will land.

This is the fastest-moving area of Vietnamese IP law and the one where online information is most out of date — most circulating articles predate the framework that took effect on 1 April 2026.

Three settled principles

AI is not a subject of IP rights. An AI system cannot be named as author, cannot own rights, cannot be an inventor. That closes the "machine author" debate in Vietnam.

Fully automatic AI output is not protected. Without human involvement the output is not protected as a human work or invention. It belongs to no one — not even the person who pressed run.

A human using AI as a tool can still be the author. Where the user makes a substantial creative contribution — conception, direction, selection, editing of the output — they may be recognised as author or inventor. Decree 100/2026/ND-CP addresses how industrial property rights arise where subject matter was created using an AI system.

The unsettled boundary

"Substantial creative contribution" is not quantified. Between the clear extremes — a short prompt taken verbatim, versus AI as one stage in a long creative process with dozens of iterations and manual reworking — lies a wide grey zone.

In that grey zone, what decides matters is what you can prove. Which is the most practical reason to keep a record of your process: save prompts and intermediate outputs; keep sketches and manual edits before and after; use tools with automatic version history; and note what you selected, rejected and why.

For commercially valuable work, that record is the evidence you will need in an authorship dispute.

Training data

Law 131/2025/QH15 frames the use of IP-protected material in AI training. The approach: use is not infringing where the data was lawfully published and publicly accessible, the use does not unreasonably prejudice the rights holder's legitimate interests, and — for copyright material — detailed conditions set by the Government are met.

The structure is familiar: it mirrors the three general conditions for copyright limitations and exceptions. What to watch is the Government's detailed conditions, since that is where the real boundary gets drawn.

Three practical scenarios

Using AI for marketing copy. Give a brief and use the output verbatim and it is likely unprotected — meaning you would struggle to stop a competitor copying it word for word. Write the outline, iterate, add your own material, and your position is far stronger.

Using AI to help write code. Code where you designed the architecture, chose the approach and edited the result remains your work. Fully generated code used as-is sits on weaker ground. For commercially important source, keep commit history that shows the human contribution.

Worried your content trains someone's model. The new framework sets conditions, but cross-border enforcement remains hard. The practical measures today are still technical controls and the terms of use on your own platform.

The short advice

If the work is commercially valuable and AI was involved: keep a record of the process, make sure the human contribution is real and demonstrable, and state in contracts with partners where AI was used.

Cited documents

Sources