Pennsylvania has approved a new law requiring disclosures in a specific category of advertisements using artificial-intelligence-modified or generated content: ads that create a false consumer perception about the nature, origin or authenticity of the goods or services being promoted. The distinction is consequential. The Artificial Intelligence Deceptive Advertising Act is not a general label mandate for all AI-generated material, nor does it require a disclosure simply because an advertiser used an AI tool. Its operative text ties the obligation to a knowingly, recklessly or willfully disseminated advertisement, a potentially misleading impression about the advertised offering, and the absence of a clear and conspicuous disclosure. The enacted bill text provides the controlling terms.

What the law actually says

Governor Josh Shapiro approved Senate Bill 806 on October 22, 2026. Pennsylvania’s legislative history identifies the measure as Act No. 99 of 2026. The enactment is titled the Artificial Intelligence Deceptive Advertising Act, and its own scope section says it “relates to deceptive advertising.” Those details matter because headlines describing a law on “AI-generated ads” can suggest a much broader regulatory project than the legislature enacted. The bill history records the approval date and act number, while the text defines the legal threshold.

The act defines “synthetic content” as information—including an image, video, audio clip or text—that has been significantly modified or generated by AI. But that definition does not itself trigger a label. Under section 4, an advertiser may not knowingly, recklessly or willfully create and disseminate, or cause to be created and disseminated, an advertisement using synthetic content in a manner that creates a false consumer perception about the nature, origin or authenticity of the advertised goods or services without disclosure. The provision is therefore about a specified deceptive use in advertising, not AI authorship in the abstract.

Put another way, an AI-generated image of an imaginary landscape used as obvious decoration is not automatically within the law merely because it is synthetic. The legally relevant question is whether the content is used in an advertisement in a way that creates a false perception about what is being sold or its nature, origin or authenticity, alongside the statute’s specified conduct standard. The act does not define every boundary of those concepts, so this example illustrates the structure of the test rather than predicting a legal outcome.

There is also a difference between the law’s definition of AI and everyday shorthand. The act supplies a broad machine-system definition and expressly includes generative AI. It does not, however, say that every piece of information produced by a system fitting that definition becomes regulated advertising. The advertisement and the misleading impression are central elements. A consumer-facing disclosure obligation follows only where the statutory conditions apply.

What a disclosure must do

When the rule applies, the advertiser must state that the advertisement contains content modified or generated using AI. The disclosure must appear in the same medium as the synthetic content and be displayed “in the first instance whenever” that content is presented to the consumer. To the extent technically feasible, it must be permanent or extraordinarily difficult to remove from the synthetic content. Those requirements are more specific than a generic instruction to “label AI.”

The statute elaborates what “clear and conspicuous” means. The disclosure must be noticeable, readable, understandable and, where applicable, audible. Its size, color, contrast, location, duration and audibility are all relevant. It must use the language or other interaction used by the consumer to interact with or comprehend the synthetic content, and it cannot contradict other information shown alongside it. When it qualifies or explains other information, it should appear close enough to that information to be noticed and must not be obscured.

For audio, the law calls for sufficient volume and cadence for a consumer to hear and understand the disclosure. For visual content, the text must be large and visible long enough to read and comprehend. Printed ads and promotional material, including point-of-sale displays and consumer-directed brochures, need a sufficiently noticeable type size and location, with print that contrasts against its background. These specifications make the medium part of the compliance question: a fleeting or inaudible notice may not meet the standard just because disclosure language technically exists somewhere.

Persistence is qualified by feasibility

The persistence requirement is not absolute. The words “to the extent technically feasible” qualify the obligation to make a disclosure permanent or extraordinarily difficult to remove. The legislature does not specify a technical standard, prescribe a particular metadata format, or state how that feasibility should be assessed for every platform or file type. The law instead combines a goal—durable disclosure—with a feasibility qualification that may require interpretation in concrete circumstances.

Boundaries and exemptions

Section 4 excludes content that is evidently fictional based on the content or context of the depiction; parody, satire, commentary or criticism; and content limited to text. These are statutory exceptions, not informal preferences. Their presence means the act does not establish a universal label requirement for AI-generated media, even when the material is publicly distributed.

The text-only exception is particularly notable because the definition of synthetic content expressly includes text. The definition captures AI-modified or generated text, but the exception then removes material limited to text from the disclosure subsection. As a result, the definition alone cannot be read as a mandate to label every AI-written passage in an advertisement. Whether a mixed-format advertisement containing synthetic text and other synthetic elements fits the text-only exception depends on the particular content; the statute does not spell out every combination.

The “evidently fictional” exception also leaves judgment calls. The law directs attention to the content or context of the depiction, but offers no checklist for when fiction is sufficiently obvious. The Verge’s October 24 report highlighted that uncertainty, along with the phrase “significantly modified” and the false-consumer-perception test. Those are interpretive uncertainties identified by a news outlet, not findings by a court or official guidance. The Verge’s account is useful as contemporaneous analysis, while the enacted language remains the source for what the statute actually provides.

“Significantly modified” is likewise undefined in the act. It might matter whether an AI system made a minor correction, transformed a photograph, or created the depicted product from scratch, but the text does not draw those lines. The statute’s operative deception condition narrows the practical issue: a change is not enough by itself; the content must be used in a way that creates a false consumer perception about the advertised goods or services. The law gives no statutory examples or quantitative threshold for significance.

There is a further qualifier in the enacted version: the advertiser’s conduct must be knowing, reckless or willful. News accounts that paraphrase the obligation as applying to any AI-generated ad omit an important part of the final text. The disclosure requirement is framed around that mental-state language as well as the ad’s deceptive effect. Whether a particular advertiser acted with one of those states of mind is a fact-sensitive issue that cannot be resolved from the bill text alone.

Who can enforce it

The act authorizes the Pennsylvania Attorney General or an appropriate district attorney to bring an action when the official has reason to believe a person is using or is about to use a prohibited practice and proceedings would serve the public interest. The official may seek a temporary or permanent injunction. The statute also authorizes civil actions for violations and allows a court to award damages, a civil fine not exceeding $3,000, and actual attorney fees and court costs.

This enforcement structure should not be confused with a private right to sue expressly granted to every consumer. The text names the Attorney General and district attorneys as enforcement actors in its provisions. It does not state in those provisions that an individual consumer may independently bring an action under the act. Other legal remedies may exist under separate law, and the act says its disclosure rule does not limit other recourse, but that savings language is not itself a newly created consumer lawsuit procedure.

Nor does the act establish a new licensing regime for AI tools, impose duties on AI developers generally, or create a broad monitoring agency. Its subject is advertising conduct, and its enforcement mechanisms run through state and local public officials and courts. That limited design is part of why it is best understood as a consumer-advertising measure rather than a comprehensive AI governance statute.

Legislative timeline and effective date

Senator Nick Pisciottano introduced SB 806 on June 3, 2025. The Senate passed its final version on June 10, 2026, by a 48–2 vote. The House approved it on October 7, 2026, by 163–40. The bill was signed in both chambers on October 20, presented to the governor on October 21, and approved on October 22, according to the legislature’s history page. The bill was designated Act No. 99 of 2026.

The effective-date clause says the act takes effect in 60 days. Because the approval date is October 22, the statutory interval—not the headline date—is what determines when the law becomes effective. Counting 60 days after enactment points to December 21, 2026, subject to the applicable rules for computing statutory time. The legislation itself does not print a calendar effective date in section 6; it specifies the 60-day interval.

The news coverage arrived promptly after approval. The Verge published October 24, and Mashable published October 25. Their timing supports treating this as a fresh development for an October 23–26 coverage window. Mashable’s report confirms the enactment was being discussed publicly, but its headline and some broad language imply a wider net than the final act’s exceptions and conditions support.

Practical questions the statute leaves open

The statute’s design creates several questions for advertisers. The first is how to assess whether synthetic material risks a false perception about a product’s nature, origin or authenticity. An image that makes a real product look more attractive may raise different issues from an image that depicts features the product does not have. The law does not offer a catalog of examples, and this article does not treat either scenario as automatically lawful or unlawful.

The second question is how to implement disclosures across formats and distribution channels. A video may circulate as a social post, an embedded clip, a broadcast spot or a reposted segment. The act’s “first instance whenever” language and same-medium requirement point toward disclosures accompanying the content when encountered, while the permanence requirement is qualified by technical feasibility. It does not prescribe one universal placement or technical mechanism for every platform.

Third, the statute’s key boundaries may need interpretation. “Significantly modified,” “false consumer perception” and “evidently fictional” are not exhaustively defined in the text. A court or enforcement authority might eventually clarify their application, but as of the October 26 reporting window, the supplied sources provide no reported enforcement action or judicial construction. That absence should not be mistaken for proof that the law will have no effect; it means the practical boundaries remain unsettled.

Finally, the law does not itself establish that a particular advertisement is deceptive, or that AI was used in a particular campaign. It sets a rule for covered conduct. Determining whether an ad falls within it would require facts about the content, the claims consumers could reasonably perceive, how the material was made and distributed, what disclosure appeared, and the advertiser’s state of mind. The statute supplies a legal framework, not a finding about any company or existing campaign.

Frequently asked questions

Does Pennsylvania now require every AI-generated ad to carry a label?

No. The act applies to advertisements that use AI-modified or generated synthetic content in a manner creating a false consumer perception about the nature, origin or authenticity of advertised goods or services, and it includes a knowing, reckless or willful conduct standard. It also contains express exceptions.

What must a required disclosure say?

It must state that the advertisement contains content modified or generated using AI. It must be clear and conspicuous, presented in the same medium as the synthetic content, appear whenever the content is first presented to the consumer, and be permanent or extraordinarily difficult to remove where technically feasible.

Are parody and AI-generated text covered?

The statute excludes parody, satire, commentary and criticism, as well as content limited to text. It also excludes depictions that are evidently fictional based on their content or context. Mixed-format situations and the meaning of “evidently fictional” are not fully resolved by the statutory text.

Who can enforce the act, and what is the maximum civil fine?

The Pennsylvania Attorney General or an appropriate district attorney may bring enforcement actions under the act. A court may award a civil fine of up to $3,000, actual attorney fees and court costs; the law also permits injunctions and damages as specified in its enforcement section.

When does the law take effect?

It takes effect 60 days after enactment. Governor Shapiro approved it October 22, 2026, so the 60-day interval points to December 21, 2026, subject to applicable time-computation rules.

References