No filming:An AI avatar or clone yourself
    Industry

    NY synthetic performer vs Article 50

    NY §396-b is ads with fake non-identifiable humans ($1k/$5k). Article 50 is EU deepfake deployers (€15M or 3%). Different triggers.

    Versely Team••7 min read

    New York's synthetic-performer ad rule and Article 50 are not the same trigger. NY GBL §396-b (in force 9 June 2026) is an advertising statute: conspicuous disclose when an ad reaching New York consumers uses a digitally created human who is not recognisable as an identifiable natural person. Penalties $1,000 first / $5,000 subsequent. Article 50 of the EU AI Act (applied 2 August 2026) is a transparency statute: providers machine-mark; deployers label deepfakes at first exposure. Fines can reach €15 million or 3%. A file can hit both, one, or neither. Do not restack them.

    This page is the comparison. The NY-only explainer is a different post. Audio and translation exemptions are NY law skips audio and translation. Digital twins are NY law is not a digital-twin statute.

    What each statute is looking at

    NY §396-b asks whether this is an advertisement, whether it reaches New York consumers, whether the producer has actual knowledge, and whether it features a synthetic performer: a digitally created asset, made with generative AI or another algorithm, intended to create the impression of a human performance by someone not recognisable as an identifiable natural person. AP and firm notes (McDermott / Crowell / Manatt, June 2026) are the public recap layer. HQ does not matter. Reach does. Audio-only ads are exempt. AI used solely for language translation of a human performer is exempt. Ads for expressive works (film, TV, games) can be exempt when the synthetic use matches the work. Publishers and platforms hosting the ad are not the advertiser. Animals, objects, and non-human imagery are out. Commentators point enforcement at the NY AG; no private right of action in the recaps we are using.

    Article 50 asks whether you are a provider or a deployer of a generative system, and whether the published content is a deepfake — AI image, audio, or video that resembles existing persons, objects, places, entities, or events and would appear authentic — or certain public-interest text. Guidelines 20 July 2026. Provider marking grace 2 December 2026 for systems already on the market before 2 August. Deployer label at first exposure. Artistic, satirical, or fictional work may label without spoiling the work. Assistive editing is exempt. Personal, non-professional use sits outside the Act. Official icons are optional; labelling is not.

    NY is "is there a fake unidentifiable human in an ad." Article 50 is "is this authentic-looking synthetic media (or covered text) published by a deployer / marked by a provider." NY cares that it is an ad. Article 50 cares that a person would take it as authentic, ad or not.

    Where they overlap and where they split

    A photoreal generated UGC presenter in a paid unit that runs nationally:

    • NY: likely in, if the person is not a recognisable real individual, the buyer knows it is synthetic, and NY consumers can see the ad. Conspicuous disclose in the ad. $1k / $5k per the statute.
    • Article 50: likely in for EU viewers, because a realistic person who was never filmed is a deepfake-shaped file. First-exposure label. Provider should have machine-marked. €15M / 3% is the Act's cap, not a NY conversion.

    A photoreal generate in organic social, not an ad:

    • NY: out. This is an ad rule.
    • Article 50: still in if it is a deepfake to an EU audience. Organic is not an exemption.

    A generated background or product turn with no fake human:

    • NY: out. Not a synthetic performer.
    • Article 50: maybe, if the place or event would appear authentic. A clearly impossible scene is a weaker case.

    A digital twin of a named creator:

    • NY: out of §396-b's definition (not an unidentifiable performer). That is likeness and right-of-publicity, a different problem.
    • Article 50: still a deepfake if the resemblance would appear authentic. Label. Get the grant. Do not treat NY's gap as an EU gap.

    Voice-only ad, no picture of a person:

    • NY: exempt as audio-only.
    • Article 50: audio can still be a deepfake if it resembles a real person and would appear authentic. Different result.

    Translation-only AI on a real human's performance:

    • NY: exempt when that is the sole AI use.
    • Article 50: still a deployer question if the result is a deepfake-shaped file; assistive cases differ. Do not assume the NY carve-out.

    Obvious animation, no realistic human:

    • NY: weak or out (no impression of a human performance that reads as a person).
    • Article 50: weak or out on resemblance plus authenticity.

    Penalties are not interchangeable. Do not quote $5,000 as an EU number. Do not quote 3% as a NY number. Third-party USD here is the NY statute, not a Versely price.

    What a team should put on the brief

    One row per destination, not one global "AI legal" tick.

    1. Is it an ad that can reach NY? If yes, look for a non-identifiable fake human. If present, conspicuous disclose in the ad. Keep the file out of the audio-only and translation-only buckets unless it truly is those things.
    2. Is there an EU audience and a realistic person, place, or event? If yes, deployer label at first exposure. Use the Article 50 disclosure template. Pick an official icon if you use one.
    3. Is the face a real identifiable person? NY §396-b is the wrong statute. Likeness paperwork is the right pile. Article 50 may still want a label.
    4. Platform toggles. YouTube, TikTok, Meta. Contract. Not NY, not the Act.

    The synthetic media disclosure glossary is the umbrella term. Article 50 is the EU name. NY is §396-b. A UGC video generator does not choose the statute. Veo 3.1 does not either.

    If you need both a NY-conspicuous line and an EU first-exposure line on the same timeline, sign up and burn the stricter placement so one file clears both readers. First frame beats a description field in both regimes.

    Do not file this as "the US AI ad law" versus "the EU AI ad law." NY is performer-in-ads. Article 50 is deepfakes-and-marking. The overlap is a photoreal fake human in a paid unit that also hits the EU. That overlap is common. It is not identity.

    FAQ

    If I disclose for Article 50, have I satisfied NY §396-b?

    Only if the disclosure is conspicuous in the advertisement in the way NY requires, and the file is actually in NY's scope. An EU icon on an organic post is not a NY ad disclosure. An EU icon on a NY-reaching ad may be part of a conspicuous disclose — check placement, not just the existence of a mark.

    If the ad never targets NY, am I out of §396-b?

    The recaps turn on reaching NY consumers, not on HQ and not on a targeting checkbox you forgot to set. Broad or untargeted delivery on a national platform is the conservative yes. Targeting away from NY is an operational question for counsel, not a slogan.

    Which fine is larger?

    The published caps are different instruments. NY: $1,000 then $5,000. Article 50: €15 million or 3% of worldwide turnover. Do not convert them into one scare number. Scope the file first; the cap that can attach is the one whose trigger you met.