AI Regulation Tracker  /  Binding statute

Saudi Arabia Adds an AI-Training Text and Data Mining Exception to Its New Copyright Law

On February 13, 2026, Saudi Arabia published a new Copyright Law in the Umm Al-Qura Official Gazette. Article 26 lets AI developers reproduce lawfully published copyrighted works to train AI systems without the author's permission and without paying the author, subject to three conditions. The law takes effect roughly 180 days after publication, in or around mid-August 2026, with implementing regulations still pending.

By Anthony Guerriero  ·  The Leveraged Years Regulatory Research  ·  Published July 19, 2026  ·  Last updated July 19, 2026

Saudi Arabia has quietly done something that lawmakers in the United Kingdom, the European Union, and the United States are still arguing about. It has written a text and data mining exception for AI training directly into statute. The new Copyright Law appeared in the Umm Al-Qura Official Gazette on February 13, 2026, issued by Royal Decree, and it comes into force roughly 180 days after publication, which puts the effective date in or around mid-August 2026. Some analyses give the date as August 1, 2026. The exact day matters less than the fact that this is now settled law with a near-term trigger, not a proposal.

What Article 26 actually does

Article 26 sits inside the law's list of permitted uses, the acts that do not infringe copyright. The new addition is the one that matters for AI. It permits reproduction of copyrighted works for the purpose of developing artificial intelligence products and algorithms, and it does so without requiring the author's permission and without requiring compensation to the author. In the words the analyses draw from the text, the exception covers "reproduction of original works without the author's permission and without compensation for the purposes of developing artificial intelligence products and algorithms."

Read that carefully, because it is a real shift. In most copyright systems, feeding a protected work into a training dataset is the kind of copying that would normally need a license. Saudi Arabia has decided that this copying, done to train AI, is simply not an infringement in the first place, provided the conditions are met. There is no opt-out mechanism for authors written into the exception, and there is no statutory royalty. Where the exception applies, the author's right to say no and the author's right to be paid both fall away.

The three conditions

The exception is not unconditional. Three requirements sit on top of it, and each one is a place where a dispute could later turn.

• Lawful publication. The work being copied has to have been lawfully published. Pirated or unpublished material does not qualify.
• Lawful acquisition. The copy the developer works from has to have been obtained lawfully. Scraping behind a paywall or from an unlawful source is not covered.
• Copying limited to what is necessary. The reproduction has to be limited to what is necessary for the intended purpose. A developer cannot use the exception as a blanket license to hoard content beyond what the training actually needs.

Analyses of the law also note that the general limits familiar from international copyright doctrine apply alongside these conditions, so a use that conflicts with the normal exploitation of the work or unreasonably prejudices the rights-holder's legitimate interests can still be a problem. The precise meaning of "what is necessary for the intended purpose" is exactly the kind of phrase that the implementing regulations, and later enforcement, will have to fill in. This is not a blanket AI-training safe harbor: three statutory conditions apply and SAIP implementing regulations are pending.

Binding, but not yet fully drawn

I want to be precise about status, because this is a statute and it deserves to be described as one. This is binding law passed by Royal Decree, not a consultation, not guidance, and not a court opinion. When it takes effect, the exception is the operative rule in Saudi Arabia.

At the same time, two things are still open. The law is not yet in force as of this writing, since the roughly 180-day window runs to about mid-August 2026. And the implementing regulations from SAIP, which are expected to accompany the law coming into force, had not been issued when this was written. Those regulations are where the ambiguous edges get defined, including the scope of the necessity condition and the treatment of AI-generated output. So the direction is fixed and binding, while some of the operational detail is still pending. Both of those statements are true at the same time, and anyone relying on the exception should track the regulations rather than the headline alone.

Why this matters for US professionals

Saudi Arabia is not the jurisdiction whose copyright law governs a work created and published in the United States. But this statute reaches US interests in two direct ways, and IP counsel should treat both as live.

First, if you represent US rights-holders, understand that inside Saudi Arabia this exception strips away the two levers you normally rely on for training use. Within the scope of Article 26, a US author's lawfully published work can be reproduced for AI development in Saudi Arabia without anyone asking permission and without anyone paying. There is no opt-out to file and no royalty to collect on that use. That does not change US law, and it does not license conduct that happens on US soil, but it does mean the objection-and-payment model you assume at home does not travel to this market.

Second, if you advise US AI developers, this is a genuine safe harbor worth understanding. A developer building or training models under Saudi law now has an express statutory basis for reproducing lawfully published, lawfully acquired works for training, without the licensing overhead that dominates the US fair-use fights and the European opt-out regime. That has real consequences for where training work gets located and how data-sourcing agreements get structured. It also puts a premium on documenting the two provenance conditions, lawful publication and lawful acquisition, because those are the facts a developer would need to prove to stay inside the exception.

For licensing teams on either side, the takeaway is that a KSA nexus changes the negotiating baseline. A rights-holder cannot assume a training license is required for uses that fall inside Article 26, and a developer should not assume a US or EU analysis carries over. This is a distinct legal regime with its own answer, and it is now on the books.

What to do now

Map any exposure to Saudi Arabia. If your client licenses content into the region, trains models there, or partners with Saudi entities, put Article 26 on the checklist. For rights-holders, adjust expectations: the training-use objection and payment rights you rely on elsewhere do not apply within the exception, so protection has to come from the provenance conditions and from contracts, not from a statutory veto. For developers, build the documentation now: keep clear records that the works used were lawfully published and lawfully acquired, and keep the copying scoped to what the training actually needs. And watch for the SAIP implementing regulations, because the meaning of the necessity condition and the treatment of AI output will be set there, not in the headline of the statute.

Browse the full AI Regulation News tracker

Informational analysis for working professionals, not legal advice. Confirm how the Saudi Copyright Law and its implementing regulations apply to your situation with qualified counsel in the relevant jurisdiction.