You labeled the article. That's not the same thing as marking it.
A team I talked to recently was confident they had content labeling handled: every AI-written article on their site carries a small “AI-generated” badge next to the byline. They do. What they don't have, and hadn't realized was a separate requirement, is anything that marks that same article in a way software can detect once it leaves the page it was published on. Those are two different duties under Article 50, and shipping one doesn't get you the other.
I wrote about the visible badge itself in an earlier post on content labeling, and that post holds up, but it was really only about Article 50(4): a human-visible label on AI-generated text published to the public. It didn't get into Article 50(2), the machine-readable marking duty, mostly because at the time it wasn't the part teams were getting wrong. It is now, and it's worth its own post.
For the full obligation, see the longer Article 50 explainer. The short version of the distinction: Article 50(4) is about what a reader sees. Article 50(2) is about what a machine can detect, independent of whether a human ever looks at the surrounding page at all. A badge sitting in the HTML next to an article satisfies the first. It does nothing for the second, because it isn't part of the content, it's markup sitting beside it, and it stays behind the moment the text does anything a badge can't follow: gets copied into an email, pasted into another CMS, pulled into a PDF, or read out through an API that returns the article body and nothing else.
Images, oddly, get this mostly for free right now. A real watermark baked into an image's own pixel data travels with the file by definition, so a mark that satisfies the visible-label duty and a mark that satisfies the machine-readable duty end up being close to the same thing for images. Text doesn't get that shortcut. You can't bake a machine-readable mark into the words themselves without changing the words, which defeats the point of publishing them, so whatever solves this for text has to be something else entirely, something that travels with the content as data rather than as pixels.
That “something else” is generally signed provenance metadata attached to the content, structured data that says where it came from, in a format a machine checks rather than a caption a person reads. It's a real gap right now, including in Disclarion's own SDK: the positioned HTML badge it returns for text handles the visible Article 50(4) label well, and the pixel watermark handles both duties for images, but text-content provenance in that stricter machine-readable sense isn't shipped yet. It's next on our roadmap, and I'd rather say so plainly than let the existing badge imply more coverage than it has.
There's also a date attached to this specific gap. Our reading of the Act, as amended by the Digital Omnibus, is that generative systems already on the market before 2 August 2026 got a grace period on the machine-readable-marking piece of Article 50(2) specifically, running to 2 December 2026. That's not legal advice, and worth confirming against your own counsel's reading rather than a blog post, but it's the date on our own regulatory timeline for exactly this reason: it's closer than the more-discussed 2027 high-risk deadlines, and it applies to more products, because it's not limited to high-risk systems at all.
None of this means the visible badge was wrong to ship first. A reader-facing label is the more obviously necessary piece, and getting it in front of people matters. It just isn't the whole obligation, and “we already label our AI content” is the kind of sentence that's true and incomplete at the same time. Worth checking which half of Article 50's content duties your own labeling actually covers, before an audit does it for you — and note that even a genuine editorial-review exemption on the visible-label side leaves this machine-readable duty completely untouched.
More from the blog
Your internal AI tools owe the same disclosure duty the customer-facing one does
An HR bot or IT helpdesk assistant used only by your own staff still interacts with natural persons. Article 50 doesn't carve out an exception for "internal-only."
Does the EU AI Act apply if you're not based in the EU?
"We're a US company, we don't have an EU entity" isn't the exemption it sounds like. The actual test is about who's on the other end of the conversation, not your mailing address.
Why I built Disclarion as an SDK, not a dashboard
The easier product to build is a dashboard you configure after the fact. I built Disclarion to sit inside the AI call itself instead, and it's worth explaining why.
Deepfakes get a stricter rule than an AI-written blog post
Article 50(4)'s deepfake duty isn't the same obligation as ordinary AI content labeling — it applies to image, audio, and video specifically, and the built-in exemptions work differently too.
Your chatbot isn't the only thing that owes a disclosure under Article 50.
Article 50's disclosure duty doesn't stop at chatbots — emotion recognition and biometric categorization systems carry the same obligation, and they're the two system types teams miss most often.
You named it "AI Assistant." That's not the same thing as disclosing it.
A bot's name feels like disclosure. Article 50 cares whether a reasonable user actually understood they were talking to an AI — not whether a label existed somewhere on the page.
Your editor touched it. Article 50's label exemption still might not apply.
Article 50(4) waives the visible AI-content label when a human takes substantive editorial responsibility for a piece — but a quick skim doesn't qualify, and the exemption doesn't touch the audit-trail duty underneath it.
Your AI answers the phone now. There's no modal for that.
Article 50's disclosure duty doesn't go away because there's no screen. What actually discloses an AI voice agent on a phone call, when a modal isn't an option.
An AI audit trail doesn't need to know what your users actually said.
Teams hear "audit trail" and picture a database of every conversation. Article 50's version is narrower: it has to prove disclosure and labeling happened, not repeat what was said.
Article 50 skips disclosure when it's "obvious from context." That word is doing more work than it looks like.
Disclosure isn't required when AI use is obvious from context, but Article 50 never defines obvious, and most of the cases teams assume qualify, don't.
Your chatbot needs to disclose it's an AI. Once, not every message.
Disclosing a chatbot is AI sounds simple until you build it: repeat it every message and people tune it out, show it once and later visitors never see it.
You have logs. That's not the same thing as an audit trail.
Under Article 50, an audit trail has to be insert-only, complete, and producible on demand, not a data-engineering project when a regulator actually asks.
Your AI-generated content needs a label. Here's what that actually means.
A disclaimer someone remembers to add isn't a labeling system. Under Article 50, AI-generated content needs to be identifiable as such, reliably, not as a manual step someone can forget.
EU AI Act Article 50, explained: who it affects and what it requires
Article 50 of the EU AI Act has been enforceable since August 2, 2026: who it applies to, its three obligations, and the penalties for getting it wrong.