EU AI Act Explained (in short)
If you don’t want to be fined…
This post was written by a human.
GDPR was already hard on compliance requirements, but Article 50 of the AI Act has taken another step with fines up to €15 million or 3% of your Global Annual Turnover (whichever is higher), should you not comply.
When the investigation comes, they won’t ask for just the safety of your users. If you asked ChatGPT or Claude to make a text for the post, an image or a video, you need to label it as AI-generated.
But there is an exception, Article 50 carves out an exception for AI generated text published on matters of public interest, as it won’t need to be labeled if it “has undergone a process of human review or editorial control” and a named person or organisation holds editorial responsibility for publishing it.
You don’t have to prove your content isn’t AI. For AI-written text, you have to prove a human took responsibility for it, and that can be enough to skip the label entirely.
That exception is narrow. It applies to text, not to images, video, audio, or code, those still have to be marked as artificially generated regardless of who reviewed them. But for the huge volume of AI-assisted writing published every day, marketing copy, articles, posts, it’s the difference between a labeling requirement and a documentation requirement.
What compliance looks like in practice
- For anything synthetic, audio, image, video, label it as AI-generated.
- For chatbots, disclose it up front.
For AI-assisted text specifically, the practical path is keeping an actual record: who reviewed the piece, when, and confirmation that they’re taking editorial responsibility for it. Not a policy document saying you do this, a record showing you did, for this specific piece of content.
When a regulator comes, you export and send it over, and they get the evidence they need.
What Article 50 doesn’t mean
It doesn’t mean you can’t use AI to write, generate, or assist. It means you have to be honest about it when you do, either by labeling it, or by being able to show a human reviewed it before it went out.
To be compliant
We give a simple, tamper-evident record of who reviewed a piece of AI-assisted content and when, so “we had a human review this” isn’t just something you say when a regulator calls, it’s something you can prove.
Sign up and try it out for free.