The EU AI Act and your files
Transparency rules for AI-generated content are why platforms started labelling in the first place. Here is how that connects to what is inside a JPEG - and where the connection stops.
Disclosure and metadata are not the same obligation
The transparency direction of the AI Act concerns telling people when content is AI-generated. It does not say that a specific metadata field must survive in every published file, and no platform badge is itself a legal requirement being discharged.
What happened in practice is simpler: once labelling became expected, platforms needed a cheap automatic signal, and the cheapest one available was already sitting in the file. That is why a regulatory conversation turned into a conversation about C2PA manifests and IPTC fields.
Where creators actually get caught
Almost never on genuinely generated work, where the creator knows what they made and discloses it. The friction lands on mixed work: a real photograph with one AI-assisted retouch, a real video with an AI upscale, a design where a background was removed by a model. The file records the tool; the label describes the post.
A defensible position
- Disclose genuinely AI-generated content. That is not something a tool can do for you.
- Know what your export carries before you publish it.
- Keep provenance where it has value - editorial work, client deliverables, anything with a compliance requirement.
- Remove it deliberately when it misdescribes your work, and keep the original file.
This page is orientation for creators, not legal advice. If a specific obligation applies to your work, take it to someone qualified to answer it.
Frequently asked questions
Does removing metadata put me in breach of the AI Act?
The transparency obligations in the Act are aimed at disclosure of AI-generated content, not at the presence of a particular metadata field. Removing metadata does not substitute for a disclosure you owe, and making that disclosure does not require you to keep every field a tool wrote. If your situation is a compliance question rather than a publishing one, ask a lawyer - this page is orientation, not advice.
So why does metadata come up in this conversation at all?
Because platforms use it as their automatic input. The regulatory direction pushed them to label, and the cheapest signal to label from is the file. That is how a policy conversation ends up being about container fields.
What is the practical takeaway?
Know what is in the file before it goes out. Whether the right move is to keep provenance or remove it depends on the work; guessing is the only option that is wrong in every case.
Check a file before you post it
The inspector is read-only and takes a second. If it comes back clear and a label still appears, the cause is not in the metadata -- and you will know that instead of guessing.