Article 50 EU AI Act: The Label Is Solved, the Responsibility Isn't

In two days, on 2 August 2026, Article 50 of the EU AI Act takes effect. For some of the teams I've talked to over the past few weeks, this isn't a deadline they've been preparing for since spring, it's one that just landed on their radar. If you produce and publish AI-generated content, you'll need to disclose from that date that it came from an AI. That covers blog posts, product descriptions, images, social posts: anywhere content is wholly or substantially AI-generated.

I'm not writing this because I'm a lawyer, I'm not, and this isn't legal advice. I'm writing it because the same question has kept landing on our desk at Laioutr over the past few weeks: "Does this apply to us, and what do we actually need to do?" Short answer: probably yes. And the answer is technically easier to solve than most people assume, but it doesn't stop at the technical layer.

What Article 50 actually requires

In short, without claiming legal completeness: AI-generated content has to be labelled, on two levels. Human-readable: a label a person notices when reading or viewing the content. And machine-readable: a marker that can be read and checked automatically, for instance by platforms, regulators, or other systems.

One nuance worth flagging for anyone who's had generative systems running for a while: for systems already live before 2 August 2026, there's a grace period until 2 December 2026 for the machine-readable marker specifically. The human-readable label isn't covered by that grace period: it applies from day one. The EU's standardised label for it is simply "AI".

And then there are the fines: up to EUR 15 million or 3 percent of global annual turnover, whichever is higher. Numbers you shouldn't ignore, but also not the actual reason to take this seriously.

Why this is more than a technical problem

The obvious reaction from a lot of teams: "It's a label, we'll set it with a script." And yes, the label itself is solvable. But the real question Article 50 is asking isn't technical: what actually counts as "AI-generated" at your company?

A piece of text an AI wrote start to finish, untouched by a human, clear case. But what about a draft an AI produces that an editor then rewrites by 80 percent? What about a product description built from AI-generated raw material but signed off by a person before it ships? Where exactly does your organisation draw that line, and who gets to draw it?

That's not a question software can answer for you. It's a question your team has to answer before any label gets applied. At this point, compliance isn't a feature you switch on. It's a process someone has to own and maintain.

What we've solved at Laioutr on the technical side - and what we haven't

We've already built the "AI" label into our platform: it's applied automatically, both human- and machine-readable, without anyone having to bolt on a manual step in the publishing workflow. For content produced and published through Laioutr, the technical side of Article 50 is, for the most part, already handled.

I say "for the most part" deliberately, not "fully". Because our software labels: it doesn't decide. It doesn't know whether your editor reworked an AI draft so heavily that it no longer counts as "AI-generated" under your internal definition. It doesn't supervise your editorial process. And it doesn't take responsibility for making sure the right content arrives correctly labelled in the right channel. That stays with you, as it should. A tool that claimed to take that decision off your hands entirely would be selling a false sense of security rather than actual compliance.

Four questions your team should answer now

Before you get into labels and systems, here are four things I'd clarify internally first:

First: who has the final say on how much human rework it takes before content stops counting as "AI-generated"? That decision shouldn't sit with each individual editor.

Second: is that definition documented across every content channel - blog, social, product descriptions, ads - or does it currently only exist implicitly for whichever channel you thought of first?

Third: who checks, before publication, that content meeting the definition is actually labelled correctly, and who's accountable when that check fails?

Fourth: is your generative system already live before 2 August 2026, or is it new? That determines whether the grace period for the machine-readable marker, running until 2 December 2026, even applies to you.

Any team that can answer these four questions has already done the hard part of Article 50. Everything after that is implementation.

Bottom line

At Laioutr, we can take the technical side off your plate: the label, the machine-readability, the missing manual step. What we can't and won't take off your plate is the decision about what counts as AI-generated at your company and who's accountable for it. That decision is a governance question, not a product question.

If you want to work on the technical side, talk to us. If you've still got open questions on the governance side, that's a conversation better had with your legal counsel than with us anyway.