Pennsylvania's AI Ad Disclosure Act: A Video Compliance Playbook

What Pennsylvania's AI ad disclosure law requires

Pennsylvania's Artificial Intelligence Deceptive Advertising Act (SB 806) raises the bar for AI ad disclosure on video: once it is signed, any ad that uses AI-made or AI-altered footage to mislead viewers about a product must carry a clear, on-screen label at first view. Here is the compliance playbook.

Pennsylvania's statute is the first US state law written specifically around AI disclosure in advertising rather than around synthetic performers or deepfakes. The General Assembly gave it final approval on Oct. 7, 2026, and the measure now sits on Governor Shapiro's desk; the act takes effect 60 days after enactment. Crucially, it amends the state's existing Unfair Trade Practices and Consumer Protection Law rather than standing alone, so it layers a fixed disclosure format on top of duties advertisers already owed.

The core rule lives in Section 4(a): an advertiser may not knowingly, recklessly or willfully disseminate an ad that uses synthetic content, meaning AI-generated or significantly AI-modified image, video, audio or text, in a way that creates a false impression about a product's nature, origin or authenticity, without a clear and conspicuous disclosure. Civil fines reach up to $3,000 per violation, enforced by the Attorney General or district attorneys. The definition of synthetic content explicitly includes generative artificial intelligence, so both fully generated clips and heavily retouched ones count.

Three carve-outs keep the law narrow: content that is evidently fictional from its context, parody, satire, commentary or criticism, and anything limited to text. That last exemption matters most for small teams, because ad copy and email subject lines drafted with AI fall outside SB 806 entirely, leaving realistic product and person depictions as the real exposure.

Pennsylvania is not writing in a vacuum; the broader {{link}} already reaches the EU, China and Turkey with similar disclosure duties. The practical test is narrow: the trigger is deception about the product, not the mere presence of AI, so an AI background behind a real product shot is a different case from an AI product render that flatters the item.

The creatives most exposed are the ones that were already shaky, including AI 'customer' clips that drift toward {{link}}. An AI spokesperson that reads like a real person, a cloned voice pitching a supplement, or a before-and-after that was never photographed all sit inside SB 806's lane.

Pennsylvania is not writing in a vacuum; the broader synthetic media ad regulation map already reaches the EU, China and Turkey with similar disclosure duties.

The creatives most exposed are the ones that were already shaky, including AI 'customer' clips that drift toward fake testimonial liability.

The 'clear and conspicuous' standard for video

SB 806 defines what clear and conspicuous actually means for a moving ad, and the rules are medium-specific. For video, that means the label is a visible, on-screen element timed to the synthetic footage, not a legal footnote buried in the description. A visual disclosure must appear on screen long enough to read on a phone, in a size and shade that stands out, while an audio disclosure must play at a volume and cadence a listener can comprehend.

Placement is where video teams make mistakes. The disclosure must show the first time the synthetic content appears, not in a closing slate, not in a description field, not behind a tap. It must sit in the same medium as the content, which means an on-screen label for footage and a spoken label for audio, and where technically feasible it should be burned in or embedded rather than a removable caption. A caption track that a viewer must switch on fails the test, because the notice must be present in the same medium at the moment of viewing.

Both disclosures must also be in the same language as the ad, so a Spanish-language spot needs a Spanish label. That last clause matters because short-form video is exactly what opponents flagged: prescriptive formatting can overwhelm a six-second clip. The law's answer is to tie the label to the moment the synthetic content is seen, not to a fixed corner bug, which pushes teams to design disclosure into the edit rather than bolt it on after delivery.

Why a burned-in label is not enough

A burned-in label survives most re-encoding, but it does not travel with the file as data, and modern ad delivery re-derives creative after you ship it. Google Ads video enhancements, for example, are on by default and use generative AI to flip or extend an advertiser's video into new aspect ratios, so the cut you approved is rarely the cut that finally runs.

File-level provenance closes that gap. C2PA's Content Credentials attach an edit history and origin record to the asset itself, so a disclosure marker stays attached no matter how many aspect ratios a platform spins off. C2PA supports both hard binding, where a cryptographic hash ties the record to the exact bytes, and soft binding, such as an invisible watermark, that survives transcoding when the file is re-encoded. The EU AI Act already requires generative AI providers to make content identifiable, and the same instinct is moving into ad workflows: disclosure as metadata, not just as pixels.

For a video team the takeaway is to treat the two layers as complementary. Burn the visible label for the viewer; embed the provenance record for the platform and the audit trail. One is a consumer-protection step, the other is an operational one, and SB 806's where technically feasible clause is the bridge between them. Treating provenance as metadata also future-proofs the asset, because a buyer or platform can read the origin record long after the specific editing tool that made the clip has been retired.

A pre-ship disclosure checklist for video teams

The law is narrow enough that most ads clear it at the first question, so the work is mostly disciplined triage. The three exceptions, fictional, parody and text-only, remove most AI-written copy and obviously synthetic scenes, so the real work concentrates on realistic product and person depictions. Before any spot that will run in Pennsylvania, ask whether it contains AI-generated or significantly AI-modified image, video or audio; if it is text only or evidently fictional, the rule does not apply.

Then ask the sharper question: does the synthetic element shape what a viewer believes about the product's nature, origin or authenticity? An AI scene implying domestic manufacture, a synthetic expert, or a product render that differs from the real unit all cross the line. This is the judgment call the bill leaves to the team, and the safest habit is to document the reasoning in writing even when the answer is no. If none of those is true, ship; if one is, add the label.

When a label is required, use the bill's wording and its placement rules: state that the advertisement contains content modified or generated using artificial intelligence, show it at first instance, keep it in the same medium, and hold it on screen long enough to read. Save the final cut, the AI tools used, and the sign-off sheet together, because the record is the only defense if a prosecutor ever asks.

Where this fits the wider disclosure map

Pennsylvania follows a path already cut by {{link}}, which also targets machine-readable provenance at the file level. The difference is scope: California's law reaches AI-generated content broadly, while SB 806 is narrowed to deception about a product, which is why small teams will clear most spots after a single triage pass.

A label only earns its protectiveness if viewers actually register it, so the {{link}} still governs outcomes. Disclosure that is technically compliant but visually invisible does neither the consumer nor the brand any good, which is why the clear and conspicuous test is the part of SB 806 worth designing around.

The direction of travel is clear: US states are beginning to write AI ad disclosure into statute, and Pennsylvania's version is the most operational yet because it specifies placement and wording instead of vague notice. Pennsylvania is unlikely to be the last, and the pattern across states is a law that names the exact label and its placement rather than delegating those details. Teams that build the triage and the dual-layer label now will meet the next state law with a template, not a scramble.

Pennsylvania follows a path already cut by California's AI Transparency Act, which also targets machine-readable provenance at the file level.

A label only earns its protectiveness if viewers actually register it, so the perception gap still governs outcomes.