Now your AI content has to say it's AI

For three years the implicit rule of AI content was: you don't have to say. The image, the article, the chatbot answering your support ticket — none of it had to announce what it was. That rule is ending, on a date.

On June 10, 2026, the European Commission published its Code of Practice on marking and labelling AI-generated content, the practical playbook for transparency obligations that become applicable under the EU AI Act on August 2, 2026. The short version: deepfakes and AI-generated text on matters of public interest must be clearly labelled, and people must be told when they're interacting with an AI system rather than a human. The Code itself is voluntary. The obligation it helps you meet, under Article 50 of the Act, is not.

This is a turn worth understanding, because it's bigger than one regulation.

The default just flipped from "hidden" to "disclosed"

Until now, disclosing that something was AI-made was a choice — usually a marketing one, made when it helped and skipped when it didn't. The EU is flipping the default. For a large class of content, the question changes from "should we mention it?" to "we have to mention it, so how do we do it well?"

And the EU rarely stays the EU. GDPR was a European law that became the way most of the world handles privacy, because building two versions of your product — one that respects the rule and one that doesn't — is more expensive than just building the compliant one. Content labelling is on the same path. Even if you never serve a European user, the tooling, the norms, and your competitors are all about to assume disclosure is standard.

Two different things are being labelled

It's worth separating the two requirements, because they hit different builders.

The first is machine-readable marking — invisible signals baked into AI-generated media (watermarks, metadata, provenance tags) so platforms and tools can detect "this was AI-made" even after it's been shared around. That's a job for the people generating images, video, and audio at scale.

The second is human-facing disclosure — a visible label on a deepfake, a note that an article was AI-written, a "you're chatting with an AI assistant" line on your bot. That one touches almost everyone shipping an AI feature. If your support chat is a model, your user is now owed a clear heads-up that it isn't a person.

Treat it as trust, not just compliance

The easy mistake is to file this under legal and do the bare minimum. The better read is that disclosure is becoming a trust signal, and the people who do it well will benefit from doing it early. We've already watched AI erode the assumption that a confident, fluent answer is a true one; in a world flooded with synthetic text and video, "we tell you what's AI" becomes something users actively look for.

So build the label in now, plainly, before you're forced to. Tell people when they're talking to your bot. Mark what your system generates. Keep the provenance so you can prove what came from where. None of that is hard if you design for it from the start; all of it is painful if you bolt it on the week before August 2.

The bottom line

The era of silent AI content is closing. The EU set a date — August 2, 2026 — and a playbook, and the rest of the world will drift toward the same default the way it did with privacy. Disclosure is becoming table stakes.

The builders who win won't be the ones who hid the AI the longest — they'll be the ones who labelled it first and made honesty a feature. Saying "this is AI" is about to be required. Say it well, and it's an advantage instead of a chore.