AI Manipulated or Generated? And how do the fines work?

Published [August 17, 2026] · 2 min read

This post was written by a human.

Life is hard, and unfortunately EU law makes it even more complicated, because your obligations differ depending on what the AI made and how. The fines are still large with up to €15 million or 3% of global annual turnover, whichever is higher.

Generated or Manipulated?

AI Generated means the AI created the whole resource while AI Manipulated means something real that the AI has been used on. Important to note that neither term is about how much a human helped, both describe the output, not the process.

For text, the rule is narrower than people think

First is the scope. Text disclosure recording only applies to text published to inform the public on “matters of public interest”. The regular marketing copy, product descriptions or internal documents are not covered by it.

When it does apply, whether the text was AI Generated or AI Manipulated barely matters, as the exception is not about how the AI was involved, but about how a human was.
If the text goes through human review, and a named person/organisation holds editorial responsibility for publishing it, you can skip labelling the content as AI.

For text, it’s about “did a human take responsibility, and can you show it?” instead of “how much did the AI write?”

For images, video and audio, there is no such exception

There is no human-review exemption here at all. What matters is whether the result could pass as authentic. If it resembles a real person, place or event and could be mistaken for a genuine photo/recording, it needs to be marked, no matter how much a human reviewed or changed it afterwards.

An obvious illustration or a cartoon of a real person is a different case. The test is whether someone could reasonably take it for the real thing.

So if we use AI to tidy up our content, what do we do?

For text on matters of public interest, a human takes responsibility, and you keep a record proving that review happened. For image, video and audio, you label it as AI. There is no percentage in the law that explains this for us and anyone quoting you one has invented it.

The best case to pursue is creating a defensible, transparent and truthful record of your AI interactions, instead of hoping a judgement call will land in your favour.

This post is not legal advice. The grey areas above are our reading of Article 50, not settled law, and anything specific to your situation should go to qualified counsel.