Pennsylvania now requires labels on some AI-generated ads from December 21, even ads made outside the state
If your automation makes ad creative, a state law now has opinions about your render step.
What happened
On Thursday, October 22, Pennsylvania Governor Josh Shapiro signed Senate Bill 806, the Artificial Intelligence Deceptive Advertising Act. It passed the Senate 48 to 2 and the House 163 to 40, according to sponsor Sen. Nick Pisciottano's office.
According to an analysis by law firm Loeb & Loeb:
- Disclosure is required when AI-generated or significantly AI-modified content is used "in a manner that creates a false consumer perception about the nature, origin or authenticity" of what is advertised.
- The label must say the ad "contains content that was modified or generated using artificial intelligence," be clear and conspicuous, appear the first time the content is shown, in the same medium, and be permanent or very hard to remove where technically feasible.
- Exemptions cover evidently fictional content, parody, satire, commentary or criticism, and content limited to text.
- There is no knowledge requirement. Enforcement is by the attorney general and district attorneys, with fines up to $3,000 plus fees and costs.
- It takes effect December 21, 2026.
The Verge notes the law applies regardless of where an ad was made, so ads expected to air in Pennsylvania are covered. Both The Verge and Loeb point out that the trigger is vague. Loeb's practical read: when in doubt, disclose.
My take
I build content pipelines for clients in several countries, and this is the kind of rule that is easy to forget until a campaign is already live. Being based in the Philippines does not change anything if the ad runs in Pennsylvania.
What I would add to any pipeline that generates ad images, video or audio:
- A field on every asset record: AI generated, AI modified or human made, plus the model used.
- A render step that stamps the exact disclosure wording on flagged assets, in the same medium.
- A targeting check: if the campaign can reach Pennsylvania, the flag decides whether the label goes on.
- A log of who approved each asset and when.
None of this is hard to build. It is just easy to skip. This is my reading of public reporting, not legal advice, so have your client's counsel confirm the trigger.
More posts
- Anthropic's open source FlashProve can prove which model wrote a response. It is a prototype, so check your AI provider yourself for nowNov 4, 2026
- Claude's new message threads cut time to first token by about 35% on 500k to 1M token requests. Token usage stays the sameNov 4, 2026
- LangChain tested seven browser toolsets for agents. All passed 78% to 90% of tasks, but one cost about 13x moreNov 4, 2026
