New York's synthetic performer ad disclosure
Since 9 June 2026, ads reaching New York consumers with a fully synthetic human performer need a conspicuous disclosure. Scope, exemption and penalties.
Most US advertising rules about AI are not about AI. They are general deception rules that happen to apply when AI is involved: say something false in an ad and the FTC's Section 5 authority reaches it whether a human or a model produced the claim. New York broke that pattern. Since 9 June 2026, a specific fact about how an ad was made — that its human performer does not exist — triggers a disclosure duty on its own, with no requirement that anyone was misled about anything else.
That makes it the first US law aimed specifically at AI in advertising rather than at deception generally, and it changes what a compliance review has to look at. The question is no longer only "is this claim true." It is also "is there a person on screen, and is that person real."
The trigger is narrower than most summaries suggest
The law, passed as S.8420-A / A.8887-B and announced by Governor Hochul as a first-in-the-nation measure, requires a conspicuous disclosure when an advertisement that reaches New York consumers features a fully synthetic human performer.
Read that as three separate conditions, all of which have to hold:
- It has to be an advertisement. Organic content, editorial, and your own product documentation are a different question. This is an ad rule.
- It has to reach New York consumers. Not "be made in New York," not "be placed by a New York agency." Reach. For anything running on a national platform with untargeted or broad geographic delivery, assume yes.
- It has to feature a fully synthetic human performer. This is the condition people get wrong, in both directions.
That third one deserves its own paragraph. "Fully synthetic human performer" is a much tighter category than "AI was used." A generated background, an AI-written script, an AI-composed music bed, an upscaled product shot, a synthesised voice-over on a live-action plate: none of those is a synthetic performer. Conversely, a generated presenter who reads your script to camera is exactly what the rule is about, and the fact that they look stylised or obviously rendered does not obviously get you out of it. The distinguishing fact in the text is that the performer is synthetic, not that a viewer might be fooled.
The other direction matters just as much. A real, consenting person whose performance has been digitally altered is not a fully synthetic performer. That is a different legal object with a different body of rules attached to it — the SAG-AFTRA synthetic performer provisions on the union side, and state digital-replica statutes like California's AB 1836 and AB 2602 on the likeness side. Do not fold them together in a single internal policy. They have different triggers, different consent requirements and different consequences for getting them wrong.
The expressive-works carve-out, and its limit
There is an exemption for promotional material for expressive works. The commercial logic is obvious: a trailer for a film with a digitally created character, or marketing for a game whose cast is by definition not real, would otherwise carry a disclosure that tells the audience something they already know about a medium built on invention.
Two cautions about leaning on it.
First, the exemption attaches to promoting an expressive work, not to being creative. An ad for a subscription box that happens to be shot like a short film is still an ad for a subscription box. "Our creative is very cinematic" is not the argument the carve-out makes.
Second, if you sit inside an entertainment marketing team, the exemption covers your trailers and probably does nothing for the brand-partnership creative running alongside them. Those are two different ad objects, often produced by the same people in the same week, and only one of them is promoting the expressive work.
What it costs
Penalties are stated at $1,000 for a first violation and $5,000 for each violation after that. On its face those are small numbers. They are also per-violation numbers attached to advertising, which is an activity that produces a lot of discrete units.
The realistic exposure is not one fine. It is a synthetic-presenter creative concept that ships across a dozen placements and runs for a quarter before anyone in the organisation asks the question, at which point the count is not one. Treat the penalty schedule as a reason to catch this at brief stage rather than as a budget line you can absorb.
A review step that actually fits a production workflow
The good news is that the trigger is a factual property of the asset, not a judgement call about consumer perception. That makes it checkable in a way most disclosure rules are not. Add one field to your asset record and one question to creative review.
| Question at review | Answer | What it means |
|---|---|---|
| Is there a human figure on screen or in voice-over? | No | No synthetic-performer duty. Other rules may still apply. |
| Is that figure a real person? | Yes, filmed | Not this rule. Check likeness consent and, if altered, digital-replica rules. |
| Is that figure a real person, digitally replicated? | Yes | Not a fully synthetic performer. Consent and replica rules apply instead. |
| Is that figure generated, resembling no specific real person? | Yes | This is the trigger. Disclose. |
| Is the ad promoting an expressive work? | Yes | Exemption territory. Confirm with counsel before relying on it. |
Two operational notes on running that table.
Record the answer at generation time, not at review time. By the time an ad is in final review, nobody remembers whether the presenter came from a stock avatar, a digital twin or a generated character, and those three answers land in three different rows of the table above. Log it when the asset is made. A one-line note in the project file is enough.
Write the disclosure once and reuse it. "Conspicuous" is a familiar standard from the rest of advertising law and it consistently means the disclosure has to be noticeable to an ordinary viewer in the medium where the ad appears — not buried, not in a caption nobody opens, not four seconds after the presenter has stopped speaking. In practice that pushes toward a burned-in on-screen line rather than platform metadata, because a burned-in overlay survives the re-encode, the crop and the repost. Writing a disclosure line nobody scrolls past covers the copy problem; adding a text overlay to the video is the mechanical part.
If the same ad carries a paid-partnership label as well, the two disclosures stack rather than substitute, and disclosure stacking is the layout problem you now have. A synthetic-performer notice does not discharge an FTC endorsement disclosure, and vice versa.
Where this sits against everything else landing right now
New York is a narrow, ad-specific, US-state rule that arrived before the broader frameworks did. Two of those broader frameworks bit on 2 August 2026: Article 50 of the EU AI Act, and California's AI Transparency Act marking obligation as amended by AB 853, with platform-side duties following on 1 January 2027. China's labelling measures have required both a visible and a machine-readable label since 1 September 2025.
Those regimes ask different questions. The EU and California care about machine-readable marking of synthetic content generally. New York cares about one visible sentence in one type of ad. A file can satisfy either without satisfying the other, which is why general AI ad disclosure compliance is worth reading as a whole rather than jurisdiction by jurisdiction.
FAQ
Does an AI voice on a real actor trigger it?
The trigger described is a fully synthetic human performer. A real, filmed performer with a synthesised or replaced voice is not fully synthetic, so on the face of the rule this is not the provision that catches it. It is squarely inside digital-replica and consent territory instead, which is a stricter regime in several states, so this is not a route to less work.
What about an animated or cartoon character?
The rule is written around a synthetic human performer. A non-photoreal, non-human character is a poor fit for that description. This is the point where the honest answer is that the boundary between a stylised human presenter and a character is not something I can resolve from the announcement text, and it is worth asking counsel rather than guessing from a summary.
Does the disclosure have to be on screen?
The standard stated is "conspicuous." That is a well-worn term in advertising law and it centres on whether an ordinary viewer would actually notice the disclosure in the format the ad runs in. For short vertical video, that realistically means visible in the frame. Platform metadata and a line in the caption are weaker positions, and they also fail the moment the asset is reposted somewhere the metadata does not travel.
Does using a generated presenter still make commercial sense under this?
That is a performance question, not a legal one, and disclosure does change the read. What actually converts with AI avatars in ads is worth testing directly with the disclosure present in the creative rather than assuming a penalty, because the alternative — running it without the line in a market that requires it — is not the comparison you get to make.