Pennsylvania has become one of the first US states to put an AI label requirement on advertising. On Thursday, October 22, 2026, Governor Josh Shapiro signed the Artificial Intelligence Deceptive Advertising Act, according to The Verge. Any ad that "has been significantly modified or generated by artificial intelligence" must carry a label in the same format as the ad itself, whether the ad is a still image, a video, audio or text.
The law is short on definitions and long on exceptions, which makes it a useful case study for anyone who builds or buys AI-generated marketing. This post covers what the law says, where it is vague, what its opponents argued, and a practical checklist.
TL;DR: the questions people are asking
| Question | Answer |
|---|---|
| What is the law? | The Artificial Intelligence Deceptive Advertising Act, Senate Bill 806, signed October 22, 2026 |
| Who sponsored it? | Sen. Nick Pisciottano (D-Allegheny) |
| What must be labeled? | Ads "significantly modified or generated" by AI that create a false consumer perception |
| What format? | A label in the same medium as the ad; audio disclaimers must be audible, visual ones on screen long enough to read |
| Does it apply to out-of-state ads? | Reportedly yes, if they run in Pennsylvania |
| What is exempt? | Parody, satire, commentary or criticism, and evidently fictional content |
| Who opposed it? | The tech group NetChoice, in an October 4 letter |
| Penalties? | We did not find them in the coverage we could read; check the bill text |
What the law actually says
The reporting on the text comes from two places. The Verge summarizes the signed law, and MediaPost covered the October 7 floor vote. Putting them together:
- Trigger: content that has been significantly modified or generated by AI, used in an ad for goods or services.
- Materiality: MediaPost quotes the bill as mandating disclosure when AI material misleads consumers about the "nature, origin or authenticity of the goods or services being advertised." The Verge adds that the wording suggests only content that uses synthetic content in a manner that creates a false consumer perception requires a disclosure.
- Format: the disclosure has to be in the same medium as the ad. Per MediaPost, audio disclaimers must be in a "volume and cadence" that consumers can hear, visual disclaimers must stay on screen long enough to be read, and where feasible the label should be "permanent or extraordinarily difficult to remove from the synthetic content."
- Reach: The Verge says the labels are required regardless of where an ad was created, so national campaigns that also run in Pennsylvania may carry disclosures in other states.
Sen. Pisciottano framed it as an anti-scam measure. In MediaPost's quote: "Consumers deserve to know if AI is being used to create a false impression about a product or service." He added that the bill sets a "commonsense transparency standard" while still allowing businesses to use the technology.
The Verge notes that the sponsor's office did not immediately return a request for comment, and we have not seen a published enforcement guide.
Where the loopholes are
The Verge's headline sub-line is blunt: the act "leaves a lot of loopholes." The main ones:
- "Significantly modified" is subjective. Is an AI background extension significant? An AI-retouched product photo? An AI-written headline? The text does not give a threshold in the coverage we read.
- Carve-outs. Parody, satire, commentary or criticism are exempt, as is anything "evidently fictional." What counts as evidently fictional in a 15-second ad is unclear.
- The false-perception test. If a label is only needed when synthetic content creates a false consumer perception, then an AI-generated image that accurately shows a product may need no label at all. Who decides what is false, and when, is left open.
- Small advertisers. The Verge notes the law covers restaurants' crowded flyers about wing specials as much as multi-billion-dollar TV spots, which raises compliance questions for businesses with no legal team.
What NetChoice argued
The tech trade group NetChoice opposed the bill in an October 4 letter, as reported by MediaPost. Its argument was that federal and state laws already prohibit deception in ads, so "layering a parallel, AI-specific regime on top of these actively enforced laws does not enhance consumer protection." It also objected to the formatting rules, warning that prescriptive requirements "risk overwhelming short videos, small display ads, and audio spots."
Both points are worth weighing. The counter-argument from supporters is that a generic deception law requires proving harm after the fact, while a label requirement is a bright-line rule for a technology that can produce convincing fakes at near-zero cost.
How it fits the wider pattern
Pennsylvania is not alone. MediaPost reports that last month California Governor Gavin Newsom signed a law requiring ads with "synthetic" performers to clearly disclose that the digital figures were generated by AI. In 2025 Shapiro also signed a separate law criminalizing AI-powered deepfake scams, according to BABL AI, which targets fraud rather than advertising.
On explainx.ai we have tracked the same question from other angles:
- AI campaign ads: NYT counts $137M, 220 videos, 70% without a label shows how thin voluntary disclosure is in political advertising.
- Celebrity AI deepfakes and Meta and YouTube likeness detection covers the platform side of synthetic likeness.
- Spotify removing 50 million slop tracks and Etsy's AI policy update show marketplaces setting their own AI-content rules before lawmakers do.
- New York City Council's AI validation hearing is another example of local government writing AI rules with fines attached.
The direction is consistent: disclosure rather than bans. Regulators are mostly not saying you cannot use AI in ads; they are saying you cannot hide it when hiding it misleads.
Why labels are harder than they sound
The law asks for a disclosure that is, where feasible, "permanent or extraordinarily difficult to remove" from the synthetic content. That is a technical request as much as a legal one. A visible burned-in caption survives screenshots and reposts, but it can be cropped. Metadata standards such as content credentials travel with a file, but many platforms strip metadata on upload. Invisible watermarks survive more edits, yet they need a detector to read them, and a consumer cannot see them at all.
So in practice an advertiser who wants to comply will likely combine a visible on-screen or spoken label with embedded provenance data. The visible label meets the same-medium rule; the metadata helps prove the asset's history if a regulator or competitor asks. Short formats such as six-second video or a small display banner, the case NetChoice raised, will force designers to decide what the shortest acceptable label looks like. Expect early practice to be conservative: a plain "AI-generated" tag, repeated where the ad could be clipped.
What marketers should do now
None of this is legal advice, and the bill text should be read by counsel. As a working checklist:
- Inventory AI use. List every asset in a campaign that was generated or heavily edited with AI: images, voiceover, video, copy. Most teams do not know this today.
- Decide what "significant" means for you. Write an internal threshold, for example anything that changes what a product looks like or does, or any synthetic person or voice. A documented standard is better than an ad hoc call.
- Test the false-perception question. Would a reasonable viewer be misled about the product, its origin, or who is speaking? If yes, label it.
- Match the format. A text disclaimer on a video ad does not follow the same-medium rule. Plan for on-screen text in video and a spoken line in audio, at readable and audible length.
- Keep records. Store prompts, tools and edit history so you can show how an asset was made.
- Plan for multiple states. With California and Pennsylvania already acting, a single "AI-assisted" label policy across all markets is easier than a state-by-state patchwork.
What AI tool builders should do
If you ship image, video or voice generation, this law shifts some pressure to you, because your customers need to prove how an asset was made. Useful features include embedded provenance metadata, one-click visible labels in common ad formats, and export logs. If agents publish ads or content on a customer's behalf, add an approval step so unlabeled synthetic content is not posted automatically.
What we do not know yet
- The penalty structure and which agency enforces it were not in the coverage we could access.
- The effective date is not stated in the reports we read.
- No court has tested "significantly modified" or "evidently fictional."
- It is unclear how the Pennsylvania rule interacts with federal efforts on AI preemption or platform policies such as ad-network labeling.
We will add an update section when the bill text, enforcement details or a first case surfaces. If you work in advertising in the state, read the signed text on the Pennsylvania General Assembly site rather than relying on summaries.
Bottom line
Pennsylvania's act is a transparency law with big exceptions. Its real effect will depend on how regulators read "significantly modified" and "false consumer perception." For marketers, the safe course is cheap: know what is AI-made, label when it could mislead, and keep records. For everyone else, the law is another sign that AI disclosure is moving from voluntary platform policy to state statute.
Related reading
- AI campaign ads: $137M and 70% without a label
- Celebrity AI deepfakes and likeness detection
- Spotify's 50 million slop tracks removed
- Etsy's AI policy update
- NYC Council AI validation hearing
- Nicolas Cage and the Amazon AI waiver: claimed vs verified
Details reflect reporting available on October 24, 2026; check the bill text for the latest.