Pennsylvania has passed a law that applies to every advertiser, not just ads with AI performers, and other states are watching.
Short answer: Yes, if they are seen in Pennsylvania. From 21 December 2026, any ad that uses AI-generated or significantly AI-altered images, video or audio in a way that could create a false impression of what is being sold must state that it “contains content that was modified or generated using artificial intelligence”. The advertiser is liable even if it did not know AI was used. Finding those ads is an AI disclosure audit.
What does Pennsylvania’s AI ad law require?
Pennsylvania’s Artificial Intelligence Deceptive Advertising Act (Senate Bill 806, now Act 2026-99) was signed by Governor Josh Shapiro on 22 October 2026 and takes effect on 21 December 2026. It applies to any advertiser that creates and runs an ad, or causes one to be created and run, using “synthetic content”: images, video or audio significantly modified or generated by AI. Where that content creates a false consumer perception about the nature, origin or authenticity of the product or service, the ad must carry a clear and conspicuous disclosure. The law passed 48 to 2 in the Senate and 163 to 40 in the House.
- The disclosure must say the ad contains content modified or generated using AI. No exact wording is prescribed.
- It must be easy to notice and understand, in the consumer’s language, and appear the first time the AI content is shown, in the same medium.
- Where technically feasible, it must be permanent or extremely difficult to remove.
- The Attorney General or a district attorney can seek an injunction, damages and a fine of up to $3,000, plus costs. There is no private right of action.
Which ads need an AI disclosure under Pennsylvania’s law?
The trigger is a misleading impression of the product, not AI use in general. An AI-generated product shot that makes a hotel room look larger, food look fresher or a gadget look better made than it is falls squarely inside the law. So does an AI-altered real person appearing to endorse something they never endorsed. Evidently fictional content, parody, satire, commentary and criticism are exempt, and so is text-only content, which means AI-written copy on its own does not need a label.
The difficult cases sit in between. The law does not define “false consumer perception”, and legal commentators already read it two ways: one reading requires a label whenever AI depicts the product, the other only when the AI makes the product look different from what the buyer will get. During the debate, one senator asked “false as compared to what?” and the wording was not changed. Until regulators issue guidance or bring a case, brands are deciding the line for themselves, and the cautious position is to disclose wherever AI imagery shows the thing being sold.
Does Pennsylvania’s AI ad law apply to brands based outside Pennsylvania?
It applies to ads disseminated in Pennsylvania, wherever the advertiser is based. For a national or global brand that covers network and connected TV, YouTube, social feeds, retail media and the brand’s own website, because none of these stop at the state line. Geo-excluding a state with around 13 million residents from national campaigns is rarely practical, and running separate creative with and without a label doubles the versions to manage. The likely outcome for most US campaigns is that the Pennsylvania standard gets applied nationally, in the same way that California’s privacy rules became the default for many companies.
How is Pennsylvania’s AI ad law different from New York and California?
New York and California regulate who appears in an ad. Pennsylvania regulates what the ad shows about the product. That makes it much broader than either, since it applies to every advertiser rather than to one kind of ad, and closer in spirit to the EU AI Act, which also turns on whether AI content would falsely appear authentic.
| Market | Rule | Applies from | What triggers a label | Enforcement |
|---|---|---|---|---|
| Pennsylvania | AI Deceptive Advertising Act (SB 806) | 21 December 2026 | AI-generated or significantly AI-altered image, video or audio that creates a false impression of the product or service | Attorney General and district attorneys; fines up to $3,000 |
| New York | General Business Law § 396-b | 9 June 2026 | Ads containing an AI-generated human performer the advertiser knows about | Civil penalties of $1,000, then $5,000 per violation |
| California | SB 1050 | 1 January 2027 | Video or audio ads prominently featuring an AI-generated performer, including AI voices | State false advertising law, including private lawsuits |
| EU | AI Act, Article 50(4) | 2 August 2026 | Realistic AI-generated or AI-altered content resembling real people, objects, places or events | Fines up to €15m or 3% of global annual turnover |
The table covers the rules most relevant to US and European advertisers. Our guide to whether you have to label AI-generated ads covers India and South Korea as well.
Will other US states copy Pennsylvania’s AI ad law?
The signs point that way. The bill passed with large bipartisan majorities, and its sponsor, Senator Nick Pisciottano, described it as “a standard that other states can follow”. Similar bills are already moving. New Jersey’s AI Image Disclosure Act (A 5089) went before an Assembly committee in October 2026, Ohio’s HB 813 would require watermarks on AI-generated content, and a New York bill (S 6954) requiring AI tools to embed provenance data is on Governor Hochul’s desk, with a deadline to act of 31 December. Short laws with a clear consumer-protection framing tend to be copied, and this one is short.
Federal action is unlikely to stop the trend soon. Executive Order 14365, signed in December 2025, directed federal agencies to challenge state AI laws, but no court has struck one down on that basis. In Congress, the Senate removed a proposed ten-year moratorium on state AI laws by 99 votes to 1 in July 2025, and the main preemption proposal of 2026, a House discussion draft, would cover laws on how AI models are built while leaving laws on how AI is used alone. An advertising disclosure rule is a law about use.
Why “we didn’t know AI was used” is no longer a defence
Earlier drafts of SB 806 only caught advertisers acting knowingly, recklessly or wilfully. That language was removed. New York’s law, by contrast, applies only to AI performers the advertiser knows about. The difference matters more than the $3,000 fine, because most AI in brand creative is not added by the brand. It is added by agencies, production partners, retouchers and stock libraries, often as a routine step nobody thinks to mention, and it reaches the brand as a finished file with no record of how it was made.
Pennsylvania has put the risk on the party with the least visibility. The practical consequence is that a disclosure policy is not enough on its own. A brand needs to know, for every live ad, whether AI shaped what it shows, and that information currently lives in other companies’ workflows. The brands that handle this well will treat AI provenance like usage rights: something recorded at handover and checked before anything goes live, rather than reconstructed after a complaint.
How do we check our ads before 21 December?
A manual check is workable for a brand with a contained US footprint. It needs access to every ad account, a direct line into each agency and a shared spreadsheet.
- List every channel that reaches Pennsylvania: national and connected TV, paid social, search and display, retail media, out-of-home in the state, organic social and your website. Scope by where content is seen, not where media was bought.
- Ask each agency and in-house studio for an AI declaration on every live asset: which tools were used, what was generated rather than edited, and whether the product, setting or any person was created or changed by AI.
- Export every active ad from each platform’s ads manager and check the Meta Ad Library and Google Ads Transparency Center for anything else running under your brand.
- Check master files for Content Credentials (C2PA), for example with the free tool at contentcredentials.org/verify. Use masters rather than published copies, because exports and uploads often strip this metadata.
- Flag every asset where AI shows the product or service, or changes how it looks, works or performs. Set aside text-only content, obviously fantastical imagery and routine retouching that changes nothing about the product.
- Add a clear disclosure to flagged assets, shown when the AI content first appears and in the same medium: on screen for video, spoken or on screen for audio. Burn it into the creative rather than relying on a caption that can be cropped out.
- Log the asset, channels, AI use, disclosure applied, reviewer and date, and make an AI declaration a required field at every agency handover from now on.
For a few hundred live assets, budget around a week of a competent person’s time for the first pass. Most of it goes on chasing agency declarations rather than reviewing files.
When does a manual AI ad check stop working?
- Several agencies and production partners, each with different AI tools and record-keeping habits
- One AI-touched master turned into dozens of cutdowns, resizes and localised versions
- Retailers, distributors and franchisees republishing your assets on channels you do not control
- Provenance metadata stripped somewhere between the master file and the live placement
- Different rules in Pennsylvania, New York, California and the EU applying to the same campaign
- Continuous production, which makes a one-off audit out of date before it is signed off
What should we look for in software that checks ads for AI disclosure?
- Can it trace a live ad, including a local cutdown, back to its master file and the record of how that master was made?
- Does it cover organic and partner channels as well as paid media?
- Can it check each placement against the rules of the market where it is seen, and show the breaches rather than just listing assets?
- Does it work on top of the systems you already use, without moving your files?
How Medialake does it
Medialake is not a DAM and it does not store your files. It sits on top of the tools a brand already uses, fingerprints each asset back to its master, and tracks where every version is live across paid and organic channels and across markets. That is what an AI disclosure check depends on, because the question is never only whether an asset was made with AI. It is where that asset and every variant of it are running, and which rules apply there.
The check starts from what the brand knows about how its creative was made, whether from agency declarations or provenance data on the masters, and follows each AI-made master out to every live placement and local variant. Each placement is then checked against the rules of the market where it is seen, so a single view covers Pennsylvania, New York, California and the EU. Medialake took the same approach to a green-claims audit for a global footwear brand under the EU’s EmpCo rules and found 1,072 posts breaking the new requirements.
Scoping takes two to three days and deployment one to two weeks, which leaves time for a full picture before Pennsylvania’s law takes effect on 21 December 2026 and California’s on 1 January 2027. The same infrastructure tracks music licences and talent usage rights, and AI disclosure sits naturally alongside them, since all three come down to what is inside a piece of creative and where it is running.
Frequently asked questions
Do commercials in the US have to disclose AI?
In some states, yes. From 21 December 2026, Pennsylvania requires a disclosure on ads where AI-generated or AI-altered images, video or audio could mislead consumers about the product. New York already requires one for AI-generated performers, and California will from 1 January 2027. There is no federal AI ad disclosure law, but the FTC’s rules on deceptive advertising apply to AI content as they do to anything else.
Does Pennsylvania’s law cover AI-written ad copy?
No. Text-only content is exempt. The law covers images, video and audio that have been generated or significantly modified by AI, and only where they create a false impression of what is being sold.
Do we need a label if the AI image shows the product accurately?
On a strict reading, possibly not, because the trigger is a false consumer perception. The term is not defined and nothing has been tested yet, so many advisers recommend disclosing wherever AI imagery depicts the product itself. The cost of an unnecessary label is low compared with defending a case.
Does adding an AI label protect us from other claims?
No. The law expressly keeps all other legal remedies in place, so an ad that misleads remains open to action under Pennsylvania’s consumer protection law and federal rules even with a disclosure. A label tells people AI was used. It does not make a misleading claim accurate.
Is the fine per ad or per violation?
The law sets a fine of up to $3,000 but does not say whether that applies per ad, per airing or per day. For a campaign running across thousands of placements, the answer makes a large difference, and it will only become clear once the Attorney General brings a case.
Could federal law override Pennsylvania’s AI ad rules?
Not in the near term. No federal preemption law has passed, the Senate rejected a moratorium on state AI laws in 2025, and the main proposal under discussion in 2026 would leave laws on how AI is used, such as advertising disclosure, in place.
See which of your live ads use AI, and where they are running. Book a demo.


