Saudi Arabia's new Copyright Law, approved by Royal Decree No. M/169 and published in the Official Gazette (Umm Al-Qura) on February 13, 2026, took effect on August 12, 2026, replacing a copyright regime that had governed the Kingdom largely unchanged since 2003. Buried inside a broader modernization — new safe-harbor rules for platforms, a statutory work-for-hire default, and criminal fines that now reach SAR 1 million (about $265,000), quadrupled from SAR 250,000 — is Article 26, which permits reproducing lawfully published works to develop AI products and algorithms without the rights holder's permission or compensation. Legal analysts at Baker McKenzie describe it as one of the first statutory AI training-data exceptions anywhere in the Gulf Cooperation Council.
What Article 26 Actually Does
The exception is not unconditional. To rely on it, a developer's copy must originate from a lawfully published work, obtained through lawful means; the reproduction must be limited to what the AI-development purpose actually requires; and the use must not conflict with the work's normal exploitation or unreasonably prejudice the rights holder's legitimate interests, according to summaries from A&O Shearman and Baker McKenzie. That last pairing tracks the classic Berne Convention "three-step test" that shows up in fair-use and fair-dealing regimes worldwide — a sign the drafters were reaching for international legitimacy, not just local convenience.
What the statute does not resolve is left to the Saudi Authority for Intellectual Property (SAIP), which must issue implementing regulations within 180 days of the law's effective date. Those regulations are expected to define what "limited to what serves the purpose" means in practice — a proportionality test that, as Baker McKenzie's analysis notes, is inherently subjective until a regulator or court gives it edges. Also unresolved: whether commercial AI development qualifies, and whether outputs from models trained under the exception can themselves be treated as derivative works requiring separate clearance.
The Case Rights Holders Will Make
Credit where due: authors, publishers, and news organizations have a real grievance here, and it predates Saudi Arabia by years. Large language models are trained on enormous corpora of copyrighted text and images, often without a license, and the resulting products compete directly with the works that trained them — summarizing news articles, generating images in an artist's style, answering questions that used to require buying a book. A statutory exception that lets this happen without compensation, even a narrowly conditioned one, effectively transfers value from creators to AI developers by regulatory fiat rather than negotiated license. Rights holders in the EU spent years fighting for — and got, under Article 4 of the 2019 Digital Single Market Directive — the right to opt individual works out of text-and-data-mining scraping. Saudi Arabia's Article 26 currently offers no equivalent opt-out mechanism in the statute itself; if the implementing regulations don't add one, Saudi authors and publishers will have less leverage than their European counterparts, not more.
Why the Exception Still Makes Sense
That said, the alternative — requiring individual licenses from every rights holder whose work might appear in a training set — is not a realistic path to a functioning AI industry, and Saudi Arabia has made building one a national priority. 2026 has been officially designated the Kingdom's "Year of Artificial Intelligence" by Cabinet decision, and Saudi AI companies have already secured $9.1 billion in funding as the government races to diversify the economy away from oil under Vision 2030, according to Arab News. A licensing-only regime would hand a durable advantage to whichever jurisdiction moves first to clarify the legal status of AI training — exactly the dynamic that pushed Japan to enact its own broad Article 30-4 text-and-data-mining exception in 2018, well before the current generative-AI boom made the question urgent everywhere else.
The more serious risk isn't that Article 26 exists — it's that it exists without the guardrails that make similar exceptions elsewhere defensible. The EU's version comes bundled with an opt-out right and, since the AI Act, a training-data transparency obligation. Saudi Arabia's does not yet have either on the statute books. If SAIP's implementing regulations import something like an opt-out or a disclosure requirement, Article 26 becomes a genuinely balanced, internationally credible framework — proportionate regulation that gives AI developers legal certainty without simply expropriating creators. If the regulations instead leave "limited to what serves the purpose" as vague as it reads today, the exception will function as a broad shield for AI developers and a source of continuing uncertainty for everyone else, including the developers themselves, who will have built products on a legal foundation that a future SAIP ruling or court decision could still narrow.
Either way, the next eight months — the window before SAIP's regulations are due — will do more to define Saudi AI policy than the headline decree itself.