Skip to content
Insights

Cortexa Ground Truth

Explainer5 min readBy Joe Coffman
0:00 / 6:22

Tip: once it’s playing, tap any line to jump the audio there.

Labeling AI content just became law.

Three 2026 laws in the EU, California and New York now require labeling some AI-generated content. Whether yours needs one depends on what, and where.


“Do we have to label this as made by a machine?” A creative director asks it about a campaign, half-hoping for a no. For a lot of the work now, the honest answer is yes. Disclosure of Artificial Intelligence (AI) in creative work stopped being a someday concern in 2026: three governments wrote it into law, and the first deadlines have already passed. What follows is a plain-English map of what changed. A map, not legal advice; for a live campaign, bring in counsel.

What changed in 2026?

Three rules landed within weeks of each other. New York now asks advertisers to disclose a “synthetic performer” in an ad. The European Union (EU) now asks the makers of generative AI tools to mark what those tools produce, and asks the people who publish deepfakes to say so. California now asks its largest AI providers to label their output and hand users a free tool to check it. The triggers differ. The instinct underneath is the same: when a machine made it, say so. It is the logic of an ingredient label on a tin.

A disclosure is a photo credit, not a warning. It records how the work was made and leaves the buying decision to the buyer.

Three rules, in force now

The 2026 AI-disclosure laws at a glance

Does this reach your client's work?

Most agencies use AI tools. They are not the companies that build them, and that distinction carries a lot of weight here. California's rule aims at “covered providers,” which it defines as a generative AI system with over 1,000,000 monthly3 users. That is the model maker, not your studio. The tool you licensed is the party on the hook, and it already carries the metadata the law asks for.

Three questions settle most cases. What did you make? A synthetic human in an ad, an AI-generated image or voice, or text you are publishing to the public each trip a different wire. Where will it run? These are an EU rule, a California rule, and a New York rule, so the audience and the ad's reach decide which one applies. And who built the tool? If a client stood up their own model for a million-plus users, that client may be a covered provider in its own right.

What's Real?

  • From 2 August 20261, the EU AI Act's transparency rules apply. Providers must ensure AI-generated content is marked in a machine-readable format and detectable as artificially generated2, and deployers must clearly label2 deepfakes and AI text published on matters of public interest, unless a person reviewed it. Content already out before the deadline does not have to be re-labeled.
  • California's AI Transparency Act, pushed by AB 8534 to line up with the EU date, became operative the same day. It reaches a generative AI system with over 1,000,000 monthly3 users: those providers must apply a visible “manifest” label, embed a hidden “latent” disclosure in the file, and run a publicly accessible AI detection tool3 for free. Large platforms and camera makers get their own deadlines in 2027 and 2028.
  • New York's synthetic-performer law takes effect the 180th day after signing, in June 2026. Produce an ad, know it contains a synthetic performer5 (a digitally created figure meant to read as a human who is not a recognizable real person), and you must conspicuously disclose5 it. The penalty runs $1,000 for a first violation and $5,000 for each one after.
  • The words carry weight the statutes leave loose. New York does not define “conspicuous,” and the EU's “clearly” will be read case by case, with Commission guidelines1 and a voluntary code filling the gap. Early practice, rather than a bright line, is what you have to work with today.

The “so what” for anyone serving tech clients

  • Add one line to the creative intake: is anything here AI-generated, and where will it run? The answer routes the job to the right rule before the work starts, not after it ships.
  • Keep the provenance metadata on. When you export an AI-generated image, the file often already carries the “latent” disclosure the laws want, and a careless re-save can strip it. Preserving it is most of California's requirement met for free.
  • For any ad with a synthetic human, add a clear on-screen or in-copy disclosure, and get written consent before you render a real person's face or voice. That consent is a separate body of law, and a client's counsel will want to see it.
  • Put a labeling clause in the statement of work (SOW) and name who signs off. You surface the exposure; the client and their lawyers make the call. That division of labor keeps your client's name, and your own, clear of a penalty notice.

Next Reads