When an AI Actor in Your Ad Needs a New York Disclosure

If an ad displayed in New York contains a fictional, human-looking AI performer, and you created or produced that ad knowing the performer was synthetic, the ad itself must carry a conspicuous disclosure. That has been the rule since June 9, 2026, under General Business Law § 396-b as amended by S.8420-A/A.8887-B; the effective date and core requirement are in the Governor’s announcement. Where the ad was made is irrelevant; where it is shown is what counts. The law does not reach every use of AI: generated backgrounds, retouching, non-human mascots, audio-only spots, and translation-only dubbing sit outside it. Civil penalties are $1,000 for a first violation and $5,000 for each subsequent one.

For UGC work the sharp edge is the AI “creator” avatar. A photorealistic generated person delivering a testimonial-style pitch is the clearest covered case there is. A real creator whose footage got color-corrected, cleaned up, and captioned is not.

Pick the ad format, the performer, and where the ad runs; the verdict and penalty exposure update beside the controls.

The Enacted Law Is S.8420-A, Not S1228C

New York’s measure is S.8420-A/A.8887-B, which amended GBL § 396-b. It applies to commercial advertising across media: online, social, broadcast, streaming, and out-of-home. Three questions decide coverage: did a covered party create or produce the ad, did it actually know the ad included a synthetic performer, and did it put a conspicuous disclosure in the advertisement. The Senate record for S.8420-A has the bill text and history.

Do not confuse it with S1228C. The S1228C legislative page shows that related proposal sitting in the Senate Consumer Protection Committee. It is not the law in force.

The disclosure standard is not prescriptive and official guidance or enforcement interpretations may develop. Treat what follows as general information, not advice on a specific campaign.

What Counts as a Synthetic Performer

A synthetic performer is a digitally created, reproduced, altered, or modified asset made by computer using generative AI or another software algorithm, which creates the impression of a visual or audiovisual human performance and is not recognizable as an identifiable natural performer. Legal analysis separates these fictional performers from digital replicas of real people, which raise their own consent, publicity, labor, and contract questions (Manatt).

“AI was used” is not the test. The test is whether viewers are shown what looks like a human performing, and that human is nobody real.

Scenario	Treatment	Why
Fictional photorealistic AI spokesperson	Covered	Human-like performance, no identifiable real performer
AI background behind a real creator	Not covered by itself	Scenery is not a human performer
Non-human animated mascot	Outside the rule	No impression of a human performance
Retouching, cleanup, color on a real performer	Not covered by itself	Editing does not create a fictional human
Recognizable replica of a celebrity or creator	Separate review	Person is identifiable; publicity and consent issues instead
AI hand, crowd, extra, or partial-human asset	Unresolved	Sources do not settle how much performance must appear

A wholly fictional AI person selling shoes, demonstrating skincare, or reading a script to camera is the straightforward case, provided the content is commercial advertising, a covered party has actual knowledge, and no exclusion applies.

Recognizable replicas need a different review. If a brand generates a version of a real actor, influencer, customer, creator, or employee, the imperfect fit with this statute does not make the use safe. Permission, publicity rights, collective-bargaining obligations, employment terms, and contract restrictions all come into play. Creators: this is also the scenario to check against your own usage-rights clause, because a license to your footage says nothing about a synthetic version of you unless the contract addresses it.

Several boundaries were unresolved as of this review: synthetic crowds, background extras, isolated hands or faces, mixed real-and-generated performances, motion capture, conventional CGI, and synthetic voices used in otherwise visual advertisements. Escalate those rather than assume either way.

Who Carries the Obligation and What Actual Knowledge Means

The duty sits with the person or entity producing or creating the advertising content, when it has actual knowledge that a synthetic performer is included. Media outlets and platforms are excluded when they merely publish or disseminate the ad. Nothing in the law says only one participant can be responsible, so brands, agencies, creators, and production shops each need to look at what they actually did (Skadden).

Participant	Exposure	What to establish
Brand	Covered if its conduct amounts to creating, producing, or placing the ad	Who controlled the creative and what the brand knew
Agency	Covered when it develops or produces the ad	Whether staff selected, generated, edited, or approved the asset
Production studio	Covered when it creates the performance or finished ad	Tools, asset provenance, what it told the client
UGC creator	Fact-dependent; producing differs from posting supplied material	Whether the creator generated or modified the performer and knew its origin
AI vendor	Fact-dependent	Tool supplier only, or a material participant in the ad
Platform	Excluded when it only distributes finished creative	Whether it generated, edited, or personalized the ad

Commissioning or approving an ad is relevant, but neither is a freestanding trigger on its own without looking at the whole production and placement arrangement.

The platform carve-out is narrow. A network or billboard operator that accepts finished creative is differently situated from a service that generates the avatar, rewrites the video, or inserts personalized synthetic scenes. Do not extend the protection to a platform that helped build the ad.

Actual knowledge is a real condition, not an operating manual. The sources do not settle how knowledge held by one employee, creator, or vendor gets attributed across a campaign team, and they do not establish that weak vendor diligence alone equals actual knowledge. That is why records matter: what each party was told and when. Records document knowledge; they do not decide the legal standard.

For a creator this cuts both ways. If a brand hands you a generated avatar and you cut it into a UGC-style ad, you are producing, not merely posting. If you generate the avatar yourself, you know its origin.

An Ad Displayed in New York Is Covered Wherever It Was Made

Covered formats include paid social, sponsored creator and influencer content, online video and display, broadcast TV, streaming and connected TV, and digital or physical out-of-home. The law is not limited to social media or to ads produced in New York; an ad displayed in New York may be covered regardless of where it was produced.

That reading does not mean theoretical internet availability in New York is always enough. The sources do not define the threshold of targeting, delivery, impressions, or exposure that counts.

The practical version: a California brand running a nationwide paid-social campaign with an AI spokesperson cannot rely on its headquarters or production location. It has to look at whether the media buy delivers into New York. If New York cannot reliably be excluded, add the in-ad disclosure or get advice on the campaign’s facts. Media plans, targeting settings, insertion orders, geofencing records, and delivery reports show where the campaign was meant to run. They are risk controls, not safe harbors.

The Four Exclusions and Where They Stop

The enacted measure excludes audio-only advertisements, AI used solely to translate a human performer’s language, and qualifying advertising for expressive works. It also separates passive publication from creating or producing the ad (bill text).

Exclusion	Example	Limit
Audio-only ad	Radio-only commercial	Says nothing about a synthetic voice inside a visual ad
Translation-only AI	Dubbing a real performer’s speech into another language	Any other generation or alteration needs separate analysis
Expressive-work promo	Game trailer using a synthetic character as it appears in the game	Use must match the character’s use in the underlying work
Mere dissemination	Billboard operator showing supplied creative	Active creation or material editing is not protected

The expressive-work exclusion covers promotion for films, TV programs, streaming works, video games, and similar works. The performer’s use in the promo must be consistent with its use in the work. A game trailer showing a synthetic character as the game shows it may qualify; that same character pitching an unrelated product is a new use and should not be assumed exempt.

The audio exclusion covers radio-only spots. It does not answer whether a synthetic voice over video or images contributes to an audiovisual human performance.

What a Conspicuous In-Ad Disclosure Looks Like

The requirement is narrow but not detailed: conspicuous, and in the advertisement. The statute does not prescribe wording, font, size, color, contrast, placement, or on-screen duration (Skadden).

Two unofficial starting points: “Ad contains synthetic performer” and “AI-generated human in this ad”. Neither is state-approved language or a safe harbor. Plain wording still fails if viewers are unlikely to notice or understand it in the delivered format.

Do not assume any of these is sufficient on its own: a caption under a social post, a hashtag, a platform-generated AI label, file metadata, a notice reachable only by clicking through, or a label that flashes once in a video.

The conservative production practice is to put the disclosure inside the creative, test it on the smallest expected mobile placement, and keep it up long enough to be read. Then check every exported ratio, cutdown, language version, thumbnail, and platform rendering. A readable label in the 16:9 master can vanish after a 9:16 crop, compression, or a UI overlay. FTC clear-and-conspicuous concepts are a useful analogy for noticeability; they are not binding New York formatting rules for this statute.

Penalties Are $1,000 Then $5,000, and the Counting Unit Is Open

Civil penalties are $1,000 for a first violation and $5,000 for each subsequent violation. Legal analysis reports no express private right of action (Kelley Drye).

What counts as a separate violation is not settled: the sources do not say whether it is each asset, campaign, platform, publication, impression, or day. Nothing in the reported materials promises advance notice and a chance to fix an undisclosed ad before penalties are possible, so the workflow should not depend on one.

Other open questions: what presentation makes a disclosure conspicuous, how actual knowledge is attributed inside an organization, whether partial-human depictions and background extras qualify, whether conventional CGI or other non-generative algorithms are included, and what level of New York delivery creates the territorial connection.

A Pre-Publish Check for Creators and Brands

Run this before approving an AI-assisted ad.

1. Inventory every generated or algorithmically modified asset. People, voices, backgrounds, products, body parts, crowds, motion, post alterations. Not just files with an AI label on them.
2. Ask whether any asset shows a human performing. Acting, speaking, demonstrating, modeling, endorsing. Scenery or product renders alone are not that.
3. Separate fictional people from recognizable replicas. Fictional goes into the synthetic-performer analysis; a replica of a real creator, celebrity, employee, or customer goes to a rights, consent, labor, and contract review.
4. Document who knew what. Record what brand, agency, creator, studio, and vendor disclosed about each asset. Put a direct AI-use question in creator briefs, agency scopes, production intake forms, and delivery checklists.
5. Test the exclusions honestly. Genuinely audio-only? AI used solely for translation? Expressive-work promo consistent with the work? Do not stretch an exclusion over unrelated creative changes.
6. Apply the disclosure and check the delivered formats. Mobile crops, short cutdowns, streaming versions, paid-social placements, out-of-home layouts, not just the master.
7. Keep the evidence. Approved asset, vendor information, production notes, media plan, disclosure decision, and screenshots or recordings of the ad as it actually ran.

Contracts can back this up with AI-use notification, asset provenance, rights clearance, responsibility for disclosures, and cooperation if a campaign is questioned. Those clauses allocate risk between the parties; they do not remove an obligation the statute puts on you.

Creators should not assume the brand, agency, or platform handled the disclosure when the creator helped produce the ad. Brands should not assume a creator’s caption or a platform label settled it without checking the delivered ad and who did what.

Escalate to counsel when a campaign includes a recognizable replica, an uncertain partial-human asset, disputed knowledge between vendors, a platform that helped generate the creative, or New York delivery you cannot pin down.