The AI regulation is here: what it means for publishing and literary creation

Today, 2 August 2026, Article 50 of the EU Artificial Intelligence Act comes into application — and unlike other dates within this law, several of which have been pushed back along the way, this one arrives with real enforcement power from day one: Spain’s AI supervisory authority, AESIA, can now act.

What it actually requires

The rule requires labelling two types of content, no more: deepfakes — manipulated image, audio or video that appears to depict something real — and AI-generated text on matters of public interest that has not undergone human review or carries no clear editorial responsibility.

Why every field will experience this differently

This is worth stating plainly: there is no single way to “comply” with this law, because it doesn’t affect every sector equally. A news outlet publishing AI-generated stories faces a different risk than a pharmaceutical lab documenting trials, or a marketing agency generating advertising copy. Each discipline has its own point of friction with the rule, and it’s worth being wary of any generic guide promising a single answer for everyone.

This piece stays within the ground we actually know: publishing and literary creation.

What it means, specifically, for those who write and publish

For authors, publishers, cultural magazines and institutions that publish content, the heart of the rule is the exception: AI-assisted text that has undergone genuine human review, with a person taking editorial responsibility for what is published, falls outside the labelling requirement. This isn’t a loophole — it is, quite literally, the recognition that editorial judgement remains what distinguishes publishable text from text that isn’t (there is a rule — label it — and there is a way out of that rule: human review plus editorial responsibility).

This has concrete practical consequences for the sector:

• For authors who use AI within their writing process — for research support, to work through a block, or to explore structure — what matters legally isn’t whether they used the tool, but whether the final result passed through their own judgement before publication.
• For publishers and magazines receiving content from external contributors, this is a good, natural reason to ask how a text was generated — not out of suspicion, but because the answer determines whether it needs a label.
• For those offering editorial services — editing, proofreading, curation — this moment reaffirms the value of a craft long taken for granted: human review isn’t an optional step in the age of AI, it is, increasingly, the line between what’s publishable and what isn’t.

What remains unresolved

The European Commission’s interpretive guidance accompanying this rule is being published in real time, without years of doctrine behind it. It also isn’t yet clear what level of documentation will be enough, over time, to demonstrate genuine human review under a future inspection — nor how strict Spanish enforcement will be in practice. And there is a reform underway, the so-called Digital Omnibus, which has already delayed other parts of this same regulation once. None of this is settled yet.

The question that actually matters

For any organisation, publisher or author using AI in their process, the question this date leaves on the table isn’t “do I have to stop using AI?” — it’s can I show that someone, by name, reviewed this and took responsibility for it? If the answer is yes, you already comply. If it isn’t clear, that’s exactly the kind of diagnostic worth resolving now, calmly, before it needs resolving under pressure.