# Don't want your content used to train AI? Do these four things

Source: https://luatbanquyen.com/en/posts/opting-out-of-ai-training · Updated: 2026-11-16

Since April 2026 rights holders can reserve their works against AI training. But a reservation only counts when it is clear, public and machine-readable.

Decree 134/2026/ND-CP added to Decree 17/2023 a mechanism many creators have waited for: a **reservation right** (Article 37b). Authors and rights owners can refuse the use of their text and data for research, testing and AI training. Full legal background in [AI and copyright under the 2026 framework](/insights/ai-and-copyright).

The key point: the default is that lawfully published, publicly accessible data **may be used**. You must reserve actively, and the reservation must be clear and public — machine-readable rights information, technological measures, or a declaration through a collective management organisation. A "copyright …" line in the footer is not enough.

## 1. Say it in your terms of use

Add a dedicated clause stating that the site's content may not be used for research, testing or training of AI systems, including text and data mining, without written permission. Cite Article 37b of Decree 17/2023/ND-CP (as amended by Decree 134/2026/ND-CP) and date the clause.

## 2. Express it in machine-readable form

This is the step most people skip, and the one that matters most — crawlers do not read terms of use.

- Block AI-training crawlers in `robots.txt` by the user agents AI developers publish.
- Add text-and-data-mining reservation metadata (a meta tag or policy file under the TDM-reservation conventions in use) site-wide.
- For images, music and video, embed rights-management information in file metadata (IPTC, XMP, ID3) — and check your platforms don't strip it. Removing or altering rights-management information is penalised under [Decree 341/2025](/enforcement/copyright-administrative-fines).

## 3. Keep evidence of when you reserved

A reservation works from the moment you express it. Keep copies of `robots.txt`, the terms page and metadata as of the change — timestamped screenshots, a web-archive capture, or a [bailiff's record](/dich-vu/lap-vi-bang-ghi-nhan-xam-pham) for high-value content. See [Gathering evidence](/enforcement/preserving-evidence).

## 4. Know the limits

A reservation cannot claw back data collected **before** you made it, and it does not enforce itself: a crawler that ignores `robots.txt` still gets the content, and suing a foreign developer is hard. Its real value is a **legal basis** — without one you have little ground to object to the use of public data; with one, use against your wishes becomes something you can pursue.

If you **build** AI, the same rules cut the other way: respect reservations, keep technical records and training data, and prepare for royalty obligations when you exploit the result commercially.

## Related legal articles

- [AI and copyright under the 2026 framework](/en/insights/ai-and-copyright)
- [Taking down infringing content online](/en/enforcement/online-takedown)
- [Preserving evidence before you act](/en/enforcement/preserving-evidence)

## Related services

- [IP clauses in employment and vendor contracts](/en/services/ip-clauses-in-employment-and-vendor-contracts)
- [Hourly consultation for content creators](/en/services/hourly-consultation-for-creators)

## Sources

- [Government newspaper — Decree 134/2026/ND-CP on copyright and AI](https://baochinhphu.vn/dieu-kien-de-tac-pham-do-ai-ho-tro-tao-ra-duoc-phap-luat-bao-ho-quyen-tac-gia-102260415163902936.htm) (134/2026/NĐ-CP)

---

Reference material only; it does not replace legal advice on a specific matter.
