Do You Have to Label AI Content in 2026? 3 Platform Rules and 2 State Laws

Key Takeaways:

New York is the first US state to require it. Since June 9, 2026, General Business Law § 396-b(3) says a person who produces an advertisement "shall conspicuously disclose in such advertisement that a synthetic performer is in such advertisement, where such person has actual knowledge"
The penalty is small and the reach is not. One thousand dollars for a first violation and five thousand for every one after, applied to any company advertising to New Yorkers, which is why Amazon told third-party sellers in July 2026 to disclose AI-generated people in listing images
A synthetic performer is a generated stranger, not a deepfake of someone real. The statute defines it as an asset "intended to create the impression that the asset is engaging in an audiovisual and/or visual performance of a human performer who is not recognizable as any identifiable natural performer"
YouTube, TikTok and Meta all require a label on realistic AI content, and all three will apply one themselves if you do not. TikTok's own help pages say an auto label cannot be removed from your post
The exemptions are narrower than they sound: audio ads, translation of a real performer's language, and promotional material for an expressive work when the synthetic performer is used the same way inside the work
Hawaii was widely reported to have copied New York. The enrolled text of Act 247 contains no labeling requirement at all. It prohibits realistic AI imitations of identifiable people in advertising and lets them sue for up to $25,000 per advertisement
No label cures a fake testimonial. 16 CFR § 465.2 makes it an unfair or deceptive practice to misrepresent "that the reviewer or testimonialist exists"
Every rule here attaches to an advertisement or a post. None of them attaches to a product somebody subscribes to, which is worth knowing when you decide what your business actually sells

Somebody is about to hand you a finished ad with a model in it who does not exist. It will be cheaper than the shoot you were quoted. It will be on brand. And in one state it is now illegal to run it without saying so in the ad itself, which means the question has moved out of ethics and into production. Here is the working version of the rules, checked against primary sources on September 20, 2026. Three platform policies, two state laws that do very different things, one federal rule that no label fixes, and a deadline in California that quietly takes the choice away from you.

Do You Have to Label AI Content?

Yes, in two separate systems that do not talk to each other. If your content is realistic and generated or meaningfully altered by AI, YouTube, TikTok and Meta each require you to disclose it at upload. Separately, if your content is an advertisement and contains an AI-generated human performer, New York law requires a conspicuous disclosure inside the ad. Platform policy costs you distribution. The statute costs you money. Neither system asks whether you used AI. Both ask whether the result looks real.

Where the rule comes from	What triggers a label	What you have to do	What happens if you skip it
New York General Business Law § 396-b(3)	An advertisement containing a synthetic performer, where you have actual knowledge	Conspicuously disclose it in the advertisement	$1,000 for a first violation, $5,000 for each one after
YouTube	Realistic content generated or meaningfully altered with AI	Set "AI use" to Yes in the Attributes section at upload	A manually applied label, removal, or suspension from the YouTube Partner Program
TikTok	AI-generated or significantly edited content showing realistic scenes or people	Apply the AI-generated content label to the post	Content "may be removed, restricted, or labeled by our team"
Meta (Facebook, Instagram, Threads)	Photorealistic AI images, video or audio	Disclose at upload, or let industry signals do it	Meta attaches its own "AI info" label

What Counts as a Synthetic Performer?

A generated human who is nobody. New York's statute defines a synthetic performer as "a digitally created asset created, reproduced, or modified by computer, using generative artificial intelligence or a software algorithm, that is intended to create the impression that the asset is engaging in an audiovisual and/or visual performance of a human performer who is not recognizable as any identifiable natural performer." Read the last clause twice, because it is the hinge. The invented model in a leggings ad is a synthetic performer. A generated version of a real actor is not covered by this subdivision at all, because that person is recognizable, and a completely different body of law handles it. The exemptions are real but narrow. Subdivision 4 releases advertisements and promotional material for expressive works, including motion pictures, television programs, streaming content and video games, but only when the synthetic performer's use in the ad "is consistent with its use in the expressive work." Subdivision 7 releases audio advertisements, and cases where AI was used solely for "the language translation of a human performer." Subdivision 8 puts the obligation on the advertiser rather than the newspaper, network or billboard carrying the ad. Enforcement has started, quietly. Governor Kathy Hochul's office announced the law in effect on June 9, 2026. By late August, the New York attorney general's office told Mashable that four complaints had been filed and were under review, two of which Straight Arrow News obtained through a public records request and published: one against an apparel brand, one against a London eyewear company running social ads at New York residents. The clearest evidence that this law travels is not the complaints. It is that Amazon told third-party sellers in July 2026 that listing images and videos containing AI-generated people now need disclosures, citing the New York requirement. A state rule became a marketplace rule in about six weeks.

Which Platform Rules Actually Fire, and What Slips Through?

Realism is the trigger and the exempt list is longer than creators expect. YouTube's policy requires disclosure for content that "makes a real person appear to say or do something they didn't do," alters footage of a real event or place, or generates a realistic scene that did not occur. It explicitly does not require disclosure for beauty filters, color and lighting adjustment, background blur, captions, sharpening, upscaling, audio repair, idea generation, or cloning your own voice for voiceovers and dubs. That list is the useful half. Using a model to write your script, draft your title, generate your thumbnail or clean your audio is production assistance, and YouTube says so in writing. Generating a realistic shot you did not film is a disclosure. The line is the footage, not the workflow. YouTube also states that disclosing "won't limit a video's audience or impact its eligibility to earn money," which removes the usual excuse. Not disclosing is the thing with consequences: repeated failures can bring a manually applied label, removal, or suspension from the Partner Program. TikTok's Community Guidelines require creators to label AI-generated or significantly edited content that shows realistic looking scenes or people, and warn that unlabeled content may be removed, restricted, or labeled by TikTok. Two details in its help pages matter more than the rule. TikTok applies an automatic label when it detects Content Credentials attached by the tool that made the file, and once that auto label is on, "you won't be able to remove the label from your post." Some categories stay banned even with a label, including content that falsely shows a public figure making an endorsement. Meta takes the same shape from the other direction. Its policy attaches an "AI info" label when it detects industry standard indicators or when the uploader self-discloses, and since September 2024 content that was only edited with AI tools carries that label in the post's menu rather than on the post. Notice what all three have in common. The label is applied by the platform, on the platform's schedule, using signals the platform reads. You are a data source in that process, which is the same position creators occupy on every lever they do not own.

Hawaii Was Reported to Have Copied New York. It Did Not.

Hawaii's Act 247, signed July 14, 2026, contains no labeling requirement. The enrolled conference draft of HB 2137 prohibits knowingly publishing a realistic digital imitation of an identifiable individual without written consent when it is used in an advertisement, causes harm, or is used for fraud, defamation or harassment, and lets that person sue for injunctive relief plus "monetary damages of up to $25,000 per advertisement or recovery of actual damages." Earlier House drafts did carry a synthetic performer advertising disclosure part, worded almost identically to New York's. It did not survive conference. Several trade summaries published in late July still describe Hawaii as having enacted a near copy of the New York law, and the bill's own report title and description in the enrolled text say otherwise: realistic digital imitations, protections for individuals, civil remedies, no disclosure regime. The practical difference is the one worth carrying. New York regulates a missing sentence. Hawaii regulates a missing signature. If you use a generated likeness of a real, identifiable person in an ad, a label does not help you in Hawaii, because what you needed was that person's written consent before you published. Permission and duration are what a license actually buys, and a generative tool does not issue one.

The Federal Rule That No Label Fixes

There is one layer underneath all of this where disclosure is not a defense at all. Under 16 CFR § 465.2, it is an unfair or deceptive act for a business to write, create, or sell a consumer review or testimonial that materially misrepresents "that the reviewer or testimonialist exists," or that they used the product, or what their experience was. An AI-generated person saying they lost fifteen pounds on your supplement is a testimonialist who does not exist. An FTC spokesperson drew the line for Mashable: a synthetic performer obviously playing an actor "may be perfectly fine," while an AI-generated individual "providing a testimonial (which would necessarily be fake) or claiming to have specific expertise" may be deceptive under both the FTC Act and the reviews rule. The rule has carried civil penalties since October 2024, and the agency's published 2025 inflation adjustment put the maximum at $53,088 per violation, which is more than ten times New York's repeat-violation penalty. That is the same rule family that governs whether your last sponsored post was disclosed properly, and it has never cared how the content was produced. It cares whether a claim about a person is true.

Your business should not depend on how the next disclosure rule is written.

Send us your handle. We will come back with the product your audience is already asking you for, built in about three weeks, at $0 upfront.

See What We’d Build You →

Why Does California Matter If You Never Read a Statute?

Because California is building the machinery that applies the label without asking you. The California AI Transparency Act became operative on August 2, 2026 after AB 853 moved the date from January 1. Any generative AI system with more than 1,000,000 monthly users that is publicly accessible in California must offer a free detection tool and embed a latent disclosure in the image, video or audio it generates. From January 1, 2027, large online platforms with more than 2,000,000 unique monthly users must detect that provenance data, expose it in the interface, and must not knowingly strip it. Penalties run $5,000 per violation, and each day counts as a discrete violation. Put the three pieces in order and the trajectory is obvious. The tool stamps the file. The platform reads the stamp. The label appears whether you checked a box or not, and TikTok already says you cannot take that label off. This is the same infrastructure that decides what AI search engines will say about your product, read from the other end. Provenance is becoming a property of the file rather than a claim in a caption.

If You Produce Ads for Other People's Products

If you are a creator between 50,000 and 500,000 followers, most of your income is branded content, and a brand has started sending you generated assets to cut into your videos, then read § 396-b(3) closely. It applies to a person "engaged in the business of dealing in any property or service" who "produces or creates" an advertisement about it, with actual knowledge. Advertising lawyers have flagged that phrase as ambiguous about who in the chain owes the disclosure. Five things make the ambiguity somebody else's problem:

Ask what is in the asset. A written answer from the brand about whether any person in the footage was generated is the only way to reach "actual knowledge" honestly, in either direction.
Put the duty in the contract. One sentence assigning the disclosure obligation, and indemnity if the brand's answer was wrong, costs nothing at signature.
Flip the platform switch anyway. It is separate from the legal disclosure, it is free, and YouTube says it does not affect reach or monetization.
Never let a generated person make a claim. No testimonial, no expertise, no results. That is the federal line, and no label moves it.
Get consent for any real face. If the likeness is recognizable, disclosure is not the requirement. Written permission is.

The Part of Your Business That Never Needs a Label

Look at what every rule in this article has in common. New York regulates an advertisement. YouTube, TikTok and Meta regulate a post. California regulates the file a tool hands you and the platform that distributes it. All of it lands on the thing you make to get attention. None of it lands on a product somebody pays for every month. That is not a loophole, it is a description of two different businesses. One of them publishes content into a system whose rules are rewritten by legislatures and trust-and-safety teams on a schedule you do not set, and each rewrite adds a step to your production process. The other one charges money for something people open on purpose. The test takes one line. If every piece of content you published this year had to carry a machine-readable note about how it was made, how much of your income would need one? For most creators the answer is all of it, because content is the whole business. That is the exposure, and it is not about AI. AI is just the newest thing somebody decided your posts have to declare. A creator with 80,000 engaged followers who already sells something has every input a subscription product needs. Built by Foundry builds it: $0 upfront, revenue share, roughly three weeks from decision to a live product, the creator owns the business, and we run and maintain it forever afterwards so nobody becomes a software company by accident. That is the entire model, and it produces an asset that keeps charging while you argue with a checkbox.

The Label Is on the Pitch

Disclosure rules are a good map of where value is presumed to sit. Regulators write them where the deception happens, and platforms write them where attention is sold. Both of them are pointing at your content, because your content is what they think you own. Go check the Attributes panel before your next upload. Then go look at how much of your year depends on that panel at all.

See What We’d Build You →

Sources verified September 20, 2026: N.Y. General Business Law § 396-b and the enacted text of Senate bill S. 8420-A; the New York Governor's announcement of June 9, 2026; Hawaii's enrolled HB 2137 CD1; AB 853 amending the California AI Transparency Act; YouTube's Disclosing use of GenAI content policy; TikTok's Integrity and Authenticity guidelines and its About AI-generated content help pages; Meta's labeling approach, last updated October 23, 2025; 16 CFR § 465.2 and the FTC's 2025 civil penalty amounts; and reporting by Mashable and Straight Arrow News on the first complaints and the Amazon seller notice. This article is information, not legal advice. Current-state claims require revalidation before later use.

Do You Have to Label AI Content in 2026? 3 Platform Rules and 2 State Laws

Get Creator Revenue Insights

Do You Have to Label AI Content?

What Counts as a Synthetic Performer?

Which Platform Rules Actually Fire, and What Slips Through?

Hawaii Was Reported to Have Copied New York. It Did Not.

The Federal Rule That No Label Fixes

Why Does California Matter If You Never Read a Statute?

If You Produce Ads for Other People's Products

The Part of Your Business That Never Needs a Label

The Label Is on the Pitch

Creators who already did this

Samantha Mae

Liv Merima

Brandon Whiteleather

Curious what your app could be?

Keep reading