Pennsylvania's AI ad-label law starts Dec. 21, fines capped at $3,000

Share
Pennsylvania's AI ad-label law starts Dec. 21, fines capped at $3,000

Pennsylvania just put a disclosure requirement on AI advertising, with a 60-day clock and a penalty small enough to read as a rounding error. Both halves of that sentence matter.

Gov. Josh Shapiro signed the Artificial Intelligence Deceptive Advertising Act on Oct. 22, requiring a label on ads whose images, video or audio were generated — or significantly modified — by AI, with the requirement taking effect 60 days later, on Dec. 21. The bill is Senate Bill 806, now Act 99 of 2026 in Pennsylvania's unconsolidated statutes. It passed the Senate 48-2 on June 10 and the House 163-40 on Oct. 7; two House amendments died on 100-103 votes the day before final passage, so the Senate text is what became law. The Verge reported the signing and, fairly, that Pennsylvania is among the first states to require this.

The practical read for anyone buying or placing ads: if a synthetic person, synthetic product shot or synthetic voice is doing the persuading in Pennsylvania, the ad needs to say so — in a format a consumer will actually perceive, not a footnote. National advertisers should assume spillover: the statute does not care where the ad was produced, so a spot cut in California and aired in Philadelphia carries the same obligation, which is why compliance will likely get standardized nationally rather than geofenced.

What the law actually requires

The prohibition is narrower than "label all AI ads," and that matters. An advertiser may not create or disseminate an advertisement 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. Synthetic content is defined as information — image, video, audio clip or text — "significantly modified or generated by artificial intelligence."

Three format rules follow from the text, and they are where compliance gets concrete. The disclosure must appear the first time the synthetic content is shown to the consumer, be presented in the same medium as the content (so a video ad needs a visual label, an audio ad needs a spoken one), and be "permanent or extraordinarily difficult to remove" to the extent technically feasible. That last clause is aimed at the scrub-the-watermark problem, and it is the part that will give ad ops teams the most trouble: burned-in labels survive a platform's editing tools, metadata does not.

"Clear and conspicuous" is then defined across six subclauses: size, color, contrast, location, duration and audibility; no contradicting information; proximity to what the disclosure modifies; audio delivered at a "volume and cadence sufficient for a consumer to hear and comprehend"; visuals on screen long enough to read; print at a type size and against a background that a consumer can actually parse. Anyone who has built a six-second pre-roll will recognize the collision coming.

The act amends Pennsylvania's 1968 Unfair Trade Practices and Consumer Protection Law rather than standing alone, which is why its short title talks about unfair methods of competition and penalties instead of advertising.

The exemptions are where the coverage leaks

Three carve-outs do most of the work in shrinking the law's reach. Disclosure is not required for content that is "evidently fictional based on the content or context of the depiction," for parody, satire, commentary or criticism, or for content "limited to text." So a synthetic photo of a hamburger that never existed needs a label; the sponsored-post copy next to it, and the search ad above it, does not. Note also that the definition of synthetic content still includes text — the exemption covers an ad whose content is limited to text, not all text inside a visual ad, an ambiguity worth flagging to counsel.

"Significantly modified" is never defined beyond those two words. A background swap is presumably cosmetic; a model wearing a generated dress is presumably not. Nothing in the statute resolves the middle, and there is no rulemaking provision, no guidance requirement and no advisory-opinion process written into the act. Advertisers have roughly eight weeks to guess.

The knowledge standard also disappeared. The Senate amendment struck "knowingly, recklessly or willful" from the operative prohibition, so liability can attach to an advertiser who did not know the creative was AI-generated — and, as NetChoice notes in its testimony against the bill, the language reaches anyone who "causes to be created and disseminated" an ad, with the intermediary protection removed. The trade group's Amy Bos called the result "close to strict liability" and predicted platforms and ad networks become the natural defendants, since they are easy to identify and the bad actor may be overseas. Her broader argument is that the bill duplicates the state's existing false-advertising prohibition — deceptive ads were already illegal whether made with AI, Photoshop or a paintbrush — and that the practical outcome is over-labeling, which trains consumers to ignore labels.

That critique has teeth, and Pennsylvania's own legislators made a version of it. Rep. Chris Pielli, whose separate AI Transparency in Advertising Act passed the House 124-78 in June and is still sitting in the Senate, told The Center Square that SB 806 is "a watered-down version of our bill" and fixed on the penalty: "That's not a deterrent. They will gladly pay that. That's not a deterrent for bad actors." His bill carried a private right of action. This one does not.

Enforcement runs through the Attorney General, not consumers

Only the Attorney General or a district attorney may enforce the act, by injunction or civil action. A court may award "a civil fine that does not exceed $3,000" plus actual attorney fees and court costs. No private suits, no statutory damages, no per-impression multiplier written into the text.

Compare that with the cost side of compliance: legal review of every synthetic asset in a campaign, label design across six formats, and ad-server changes. A $3,000 ceiling means the fine is a line item, which pushes the law's real pressure onto reputational risk and whatever the AG chooses to make an example of. The fee-shifting provision is the sharper edge, since it lets the Commonwealth chase its costs — but it only bites if an enforcer decides to bring the case at all. Pennsylvania's Attorney General has not published guidance on the act, and the office did not respond to The Verge's request for comment on the bill's loopholes.

Pennsylvania is not alone in this lane, and the comparison is instructive. New York requires a conspicuous disclosure where synthetic performers appear in advertising. China already requires labels on AI-generated broadcast content, along with a human sign-off, as we covered in China orders labels and a human sign-off on every AI broadcast. Europe's AI Act builds machine-readable marking of synthetic output into the obligation itself. Pennsylvania's version is a label rule bolted onto a consumer-protection statute with a 60-day runway — the least technical and most enforceable-looking of the set.

What to watch

Three things decide whether the act means anything. First, whether the Attorney General publishes guidance before Dec. 21 that defines "significantly modified" — right now the definition is doing all the work and carrying none of the weight. Second, the first enforcement action: a case against a national advertiser would set the de facto standard overnight, while a case against a restaurant flyer would show the law works as written. Third, whether HB 95 clears the Senate in the session's remaining weeks and adds the private right of action Pielli says is the only real deterrent. If it does, Pennsylvania stops being a labeling state and becomes a litigation state, which is a different business for every brand that runs creative into it.

The deadline is the tell. Sixty days, no guidance and a $3,000 cap looks less like a deterrent than a signal that the Commonwealth wanted the principle on the books before the midterms and will figure out the teeth later. For advertisers, the cheap move is to label anything synthetic now, in the strongest format the medium allows, and treat the exemption list as a legal argument to have in reserve rather than a plan.

Does a $3,000 fine make an AI ad label real, or is the label doing the work on its own? Write to us via our contact page.

Get SiliconNoon in your inbox

One email a day with the AI stories that matter — what happened, the numbers, and why it matters. No spam, unsubscribe in one click.

Read more

Hinton wants an FDA-style approval gate for AI models

Hinton wants an FDA-style approval gate for AI models

A safety proposal with the field's most famous name behind it, a first hard look at who actually pays for consumer AI, and a new bet on squeezing more compute out of data centers that already exist. Geoffrey Hinton wants AI models to clear a regulator the way drugs clear the FDA. The Nobel laureate made the case on Tuesday's "Smart Girl Dumb Questions" podcast: "You're not allowed to just make a new drug and release it on the market. You have to convince the FDA.

Google appeals the EU's order to open Android to rival AI assistants

Google appeals the EU's order to open Android to rival AI assistants

Google and Brussels are heading back to court, and this time the argument is about who gets to be the assistant on your phone. Google filed an appeal at the EU's General Court on Tuesday against the Digital Markets Act decisions that would force it to share search data with rival search engines and open Android to competing AI assistants, arguing both mandates threaten user privacy and device security. Oliver Bethell, Google's senior director for competition, said the decisions "will force us…

Scammers are poisoning ChatGPT's answers with fake support numbers

Scammers are poisoning ChatGPT's answers with fake support numbers

A stealth security startup says it found hundreds of companies whose AI answers hand users scam contact details. The attack class is real and documented by others; the specific number is one vendor's word. Also today: light instead of copper inside AI clusters, and a small round aimed at what agents are allowed to do. Vigilance Security's researchers say they built a detector that queries ChatGPT, Gemini and Google AI Overview about hundreds of large companies and scores the sources behind…

OpenAI's mental health test scores clinicians below its own models

OpenAI's mental health test scores clinicians below its own models

OpenAI published the most rigorous mental health evaluation anyone has built for a chatbot this week. The awkward part isn't how the models scored — it's how the humans did. OpenAI released MentalHealthBench on Wednesday, an open benchmark of 1,215 synthetic mental health conversations built with more than 80 licensed psychologists and psychiatrists across 22 countries — and its own reported results contain an uncomfortable number: the clinicians scored 38.5% on it, while OpenAI's flagship…

Shenzhen's AI court assistant has handled 600,000 cases

Shenzhen's AI court assistant has handled 600,000 cases

Shenzhen put an AI assistant inside its courtrooms two years ago. It now has numbers to show for it — and foreign judges are booking flights to see the thing. Shenzhen's Intermediate People's Court says its AI-assisted adjudication system has helped handle more than 600,000 cases since going live on 28 June 2024, describing it as the world's first complete AI-assisted adjudication system and the first to span civil, administrative and criminal proceedings.