Two new pieces of legislation, one in New York, one across the EU, now require you to tell audiences when what they're seeing was made or manipulated by AI. Not ask permission. Not stop using it. Just say so. Fail to, and the fines get big, fast.
Most brands and agencies don't know these laws exist. Here's what you need to know.
New York: disclose your synthetic performers
Since 9 June 2026, any advertisement viewable in New York that features a "synthetic performer" (an AI-generated human, or human-like digital asset) must conspicuously disclose that fact.
Note the word viewable. This isn't about ads made in New York or aimed at New Yorkers. If your campaign is online, it's viewable in New York. Which means it applies to almost everything you publish.
The definition is broader than you might think. It's not just fully synthetic people. AI-generated avatars, synthetic voices, even isolated human features created by an algorithm could count. Penalties start at $1,000 per violation and rise to $5,000 for repeat offences. One limit worth knowing. The law covers performers who aren't recognisable as a real person. Digitally recreating or altering someone identifiable is a different risk entirely, governed by publicity rights, consent and contracts, not this statute.
The EU: it's not just people
From 2 August 2026, Article 50 of the EU AI Act goes further. It requires disclosure of any AI-generated or manipulated image, audio or video that could pass as real. And the EU's definition of a "deepfake" covers people, objects, places, entities and events.
Read that again. Places.
Film a real actor against a green screen, drop in an AI-generated city street, and you may have a disclosure obligation, even though every human in the frame is genuine. The audience believes they're looking at a real location. Under the Act, that belief is the problem.
The disclosure must be clear, present from the moment of first exposure, and accessible. An opening title card that can be cropped out won't cut it. The EU is developing a standardised "AI" label, with final guidelines expected imminently.
The penalties dwarf New York's: up to €15 million or 3% of global turnover.
Who carries the risk? Probably you.
Under the EU Act, responsibility sits with the "deployer": whoever used the AI system in production. That's not the AI tool provider. It's the agency or brand that made the content.
If you commission work, you need to know how it was made. If you create it, the obligation is yours. Either way, "we didn't know" is not a defence.
What to do now
Three things, starting this week:
- Audit. Review every live and in-production asset. Where has GenAI touched the work? People, voices, backgrounds, products?
- Disclose. Build clear, conspicuous AI labelling into your production and sign-off process. Make it a standard step, not an afterthought.
- Contract. Get clear with your partners on who identifies AI content, who labels it, and who carries the liability if something slips through.
The bigger point
None of this is a reason to stop using generative AI. We use it every day, and it has transformed what's creatively possible.
But the era of quietly shipping AI content is over. From August, transparency is a legal requirement, not a virtue.. The brands that treat disclosure as a mark of confidence, not a confession, will be the ones audiences continue to trust.
Tell people what's real. It turns out that's the law.
Footnote.
We are obviously not lawyers, what’s been shared above is our understanding as to the impact of these recent changes. We encourage everyone to read in depth and take action accordingly.
Sources.
Ivor Sims
Our Head of Innovation and Production got curious and dove into the research that underpinned this article.
New York Synthetic Performer
https://www.mcdermottlaw.com/insights/new-yorks-synthetic-performer-disclosure-law-what-advertisers-need-to-know/
EU AI Act - Article 50
https://artificialintelligenceact.eu/article/50/
Article Header Image
by Jesper Ryom. Taken from album Nights. Give it a listen: https://www.youtube.com/watch?v=AHvnT-IIHM0
