Pennsylvania makes advertisers label AI-generated images, video and audio in ads

What Pennsylvania just passed

Pennsylvania Gov. Josh Shapiro on 22 October 2026 signed Senate Bill 806, the Artificial Intelligence Deceptive Advertising Act, making it Act 99 of 2026. The General Assembly's bill record shows the measure passed the state Senate 48-2 on 10 June 2026 and the House 163-40 on 7 October, was presented to the governor on 21 October and approved the next day. It was one of 58 bills the governor's office listed as signed that day.

The law requires advertisers to add a clear label when an ad uses AI-generated or heavily AI-modified images, video or audio in a way that could give consumers a false impression of what they are buying. It takes effect 60 days after enactment, which puts the start date in late December 2026.

The prime sponsor is Democratic state Sen. Nick Pisciottano of Allegheny County, with co-sponsors from both parties, including Republican Sen. Frank Farry. In a statement quoted by WPXI, Shapiro called it "an important step in our effort to rein in this technology" and said "we need to get real guardrails passed into law".

What the law actually requires

The operative rule is short. Under Section 4 of the enacted text (printer's number 1808), an advertiser may not "knowingly, recklessly or willfully" create and spread, or cause someone else to create and spread, an ad that uses synthetic content "in a manner that creates a false consumer perception about the nature, origin or authenticity of the goods or services being advertised" without a clear and conspicuous disclosure.

The disclosure has to say that the ad "contains content that was modified or generated using artificial intelligence". It must appear the first time the synthetic content is shown, in the same medium as that content, and, "to the extent technically feasible", be permanent or extraordinarily difficult to remove. That last clause matters for online ads that get cropped, re-uploaded or screenshotted, because a label in a separate caption would not travel with the image.

The act defines synthetic content as an image, video, audio clip or text that has been "significantly modified or generated by artificial intelligence". It spells out what clear and conspicuous means: a size, colour, contrast, location, duration and volume that a consumer can readily notice and understand, placed next to the content it explains, in the same language the ad uses. Audio labels must be delivered at a volume and pace a listener can follow, and on-screen labels must stay up long enough to read.

What is exempt

Three kinds of content are carved out. The disclosure rule does not apply to content that is "evidently fictional based on the content or context", to parody, satire, commentary or criticism, or to content that is "limited to text". An obviously animated mascot or a cartoon spokesperson would generally fall outside the rule, as would AI-written ad copy on its own.

The trigger is also narrower than "any AI in an ad". The content has to create a false perception about the nature, origin or authenticity of the product or service. A restaurant that posts an AI-generated photo of a burger that looks nothing like what it serves is the kind of case the law targets. An AI-generated abstract background behind a real product photo is arguably not, although where that line falls will ultimately be settled by enforcement and courts rather than by the text alone.

The act also says that making a disclosure does not limit any other legal remedy. A label does not turn an otherwise false ad into a lawful one under Pennsylvania's existing consumer-protection law, which the bill record lists as the Unfair Trade Practices and Consumer Protection Law of 1968.

Who enforces it and what it costs

Enforcement sits with the state Attorney General and county district attorneys. Either can sue in the name of the Commonwealth to stop a practice with a temporary or permanent injunction, and either can bring a civil action for violations. A court may impose a civil fine of up to $3,000 and award actual attorney fees and court costs.

The act does not create a private right of action, so consumers cannot sue under it directly, and it does not set a per-impression or per-day penalty. For a large national brand, a $3,000 cap is modest. The bigger practical risks are an injunction that pulls a campaign and the reputational cost of a public enforcement action.

What the sponsor said versus what the text says

Some of the sponsor's public description goes further than the enacted text, so it is worth separating the two. Speaking to CBS Pittsburgh, Pisciottano said the rule covers images, video, audio and text that have been substantially rewritten or generated by AI, and that there could be a fine "and even potential jail time". He also said punishment would fall on AI companies rather than individual users.

The version AI Lookout read on 23 October 2026 exempts content limited to text, provides only a civil fine capped at $3,000 plus fees, and places the duty on the advertiser who creates or disseminates the ad, not on the AI tool maker. Pennsylvania's broader consumer-protection law has its own remedies, which may be what the senator had in mind, but Act 99 itself does not mention criminal penalties.

Pisciottano also told CBS that the law applies to ads shown in Pennsylvania even if they are made elsewhere, and that political ads are not covered. The text does not mention political advertising; it is written as a rule on advertising of goods and services, which is consistent with his reading. Election deepfakes would need separate rules.

How it compares with New York and other rules

Pennsylvania is not the first state in this space. New York's law on synthetic performers in ads took effect on 9 June 2026, according to Governor Kathy Hochul's office. That law is narrower in one way: it requires disclosure when an ad includes an AI-generated person, a "synthetic performer". Pennsylvania's rule is broader in subject matter, covering any image, video or audio of the product or service, but it applies only when the content could mislead.

Platforms have been moving in parallel with their own ad screening and labelling tools. Meta now uses a language model to screen ads, as AI Lookout covered in its report on Meta's LLM for flagging harmful signposting ads, and Google has opened a public SynthID detector that checks images, video and audio for SynthID watermarks. Those tools help, but they are voluntary and inconsistent across platforms. A state law makes the advertiser responsible regardless of where the ad runs.

The labelling question is also spreading beyond ads. Streaming services have started marking AI music, as in YouTube Music's labels for AI-generated songs, and courts are debating how to authenticate deepfake evidence.

What advertisers should do now

Businesses that advertise to Pennsylvanians have about two months to prepare. The practical steps are straightforward. Keep a record of which creative assets were generated or significantly edited with AI tools, including product photos retouched by generative fill. Decide, asset by asset, whether the AI content shows the product or service itself in a way a buyer might rely on, such as food, property, before-and-after results or a person giving a testimonial.

Where it does, add a label that says the ad contains content modified or generated using AI, place it on the content itself rather than only in a caption, and make sure it is legible on a phone screen and audible in radio or podcast spots. Burning the label into the image or video meets the "difficult to remove" standard better than metadata alone, though provenance metadata such as Content Credentials can sit alongside it.

Agencies and small businesses using AI image tools for menus, listings and social posts are likely to be the most affected, because they are the ones most likely to show an AI picture of the actual thing for sale.

What happens next

The law should take effect around 21 December 2026. AI Lookout found no guidance from the Attorney General's office on how it will interpret "significantly modified" or which cases it will prioritise as of 23 October 2026, and the governor's office published only a list of signed bills rather than a dedicated statement on the act.

Pisciottano predicted that either the federal government or many more states will follow. Whether that produces a patchwork of slightly different label wording, or converges on a common standard, is the question national advertisers will be watching. For now, a single sentence on the ad is enough to comply in Pennsylvania, and leaving it off risks an injunction.