Pennsylvania Governor Josh Shapiro signed the Pennsylvania AI Deceptive Advertising Act into law on Thursday, creating state rules that demand clear labels on AI-generated ads. Under the statute, also called the Artificial Intelligence Deceptive Advertising Act, any ad altered or built with AI models must carry an honest label so shoppers know software shaped the pitch. Because the rule covers any paid pitch reaching audiences in Pennsylvania regardless of where the work was done, ad agencies from New York to California now face new duties [1].
New Rules for AI-Generated Ads
The statute targets any commercial ad that “has been significantly modified or generated by artificial intelligence” and insists that the label match the medium of the ad itself. That means video clips must show on-screen text, audio clips must play spoken alerts, and print flyers must print clear words. State Senator Nick Pisciottano sponsored the bill to keep ads honest as text and voice tools spread into retail trade. His office did not return a request for comment after the bill was signed. State leaders wanted clear rules so buyers can spot computer art before they spend their money [1].
Pennsylvania is one of the first states to demand tags on AI content. Lawmakers moved fast because new apps can make realistic photos in seconds. The state wants to stop tricks early [1].
The rule covers every shop. It applies to prime-time television spots run by multi-billion-dollar companies, but it also binds small diners handing out paper flyers about their chicken wing specials. Any group running AI-generated ads must add the warning. Scale does not change that [1].
Scope in Video and Print Media
Regulators wrote the bill to follow where an ad runs rather than where a team built the files. If an agency based in California or New York makes a clip that airs on television in Philadelphia, that broadcast must carry the label. Because national media plans often send the same video reel to many states, ad makers must either build custom edits or put labels on every reel they ship. The statute demands that any message seen by Pennsylvania families carries the right tag, stopping firms from dodging local rules by working from out-of-state production offices [1].
The rule spans still photos, video, audio clips, and printed text. A brand can’t bypass the duty by picking another format. If AI made the pitch, the tag must appear right inside that ad. Format does not change the law [1].
The label must stay plain to see. Ad makers can’t hide tiny words in pale grey type at the bottom edge of a web banner or mumble a quick phrase at the end of a radio spot. Because the label must match the shape of the ad, print ads need readable text, video spots need clear titles on screen, and radio spots need spoken words. That symmetry stops hidden tricks [1].
Statutory Exemptions and Subjective Wording
Even with clear aims, the law leaves big loopholes that could cause fights in court. The text demands a tag when an ad has been “significantly modified” by AI tools. Yet the text never explains how much digital editing crosses that line. The gap is real [1].
Lawmakers created clear carve-outs for creative speech, exempting work that counts as “parody, satire, commentary or criticism” from labels. The measure also excuses anything that is “evidently fictional,” letting wild stories or cartoon humor run free. Again, what “evidently fictional” means is unclear under the text. While an ad showing a minivan flying past the moon looks fake to everyone, a clean digital photo of a cozy dining room could easily fool a hungry family into believing a fake cafe is open for dinner. State judges and consumer watchdogs will have to decide where harmless commercial art stops and unfair retail trade begins [1].
The text also says disclosures are needed when a brand “uses synthetic content in a manner that creates a false consumer perception” about goods or services. Again, that leaves questions about what qualifies as false under the law and who makes those decisions. Without fresh state guidelines, businesses face real confusion [1].
How Generative Media Reached Advertising
Pennsylvania passed the bill because generative artificial intelligence tools now reshape business. Generative AI models learn data patterns from training sets and output text, images, sound, and code from simple user prompts. The field took off during the early 2020s, powered by deep neural networks and large language models built on the transformer architecture that changed machine learning [2].
Teams in sales, marketing, screenwriting, healthcare, finance, and software development quickly adopted these systems to speed up their work. Today, text chatbots like ChatGPT, Claude, Microsoft Copilot, DeepSeek, Doubao, Google Gemini, Grok, Kimi, and Qwen write sales pitches in seconds. For graphics, text-to-image tools like DALL-E, Adobe Firefly, Stable Diffusion, and Midjourney let artists make photo-style ads without paying camera crews or renting stages. In audio and video, ElevenLabs and 15.ai clone voices, Suno and Udio make music, Google Translate and DeepL handle text, and video models like Veo, LTX, and Sora produce moving clips from brief words. These tools made synthetic ad production fast and cheap for everyone [2].
The quick spread of AI media brought real trouble. Models often invent false facts with strong confidence, a flaw called hallucinations. Scammers also use fake media for cybercrime, fake news, and deepfakes that mislead the public. In addition, large data centers draw heavy power and fresh water for cooling. Those dangers drove Pennsylvania to act [2].
Early Roots Behind Modern AI-Generated Ads
The math behind today’s AI-generated ads began long before Silicon Valley startups built web apps. Algorithmic media generation started with the Markov chain, introduced in 1906 by Russian mathematician Andrey Markov, who studied vowel and consonant patterns in the poem Eugene Onegin. By calculating transition odds between letters, Markov showed how simple math models could assemble text from training data. In the early 1970s, artist Harold Cohen built AARON, an autonomous computer system designed to paint original art. Harold Cohen showed that machines could create visual works on paper [2].
In the 1980s and 1990s, scientists built generative planning tools using symbolic AI to make manufacturing schedules and military action plans. Machine learning long favored discriminative models that sort data over generative tools that make it. That changed in 2014 when researchers created the variational autoencoder and generative adversarial network [2].
The true turning point arrived in 2017 when the paper Attention Is All You Need introduced the transformer architecture. OpenAI used that work to build GPT-1 in 2018. In March 2020, an anonymous MIT researcher shared 15.ai for voice synthesis, and OpenAI released DALL-E in 2021. Open-source coders soon built VQGAN+CLIP and DALL-E Mini, later called Craiyon. Today, as those automated creative engines power everyday sales pitches across media markets, Pennsylvania law gives buyers a simple way to know when an algorithm made the pitch [2].
- ONLINE NEWS O’Brien, T. (2026, October 24). Pennsylvania will require labels on AI-generated ads. The Verge. [Article Link]
- WEBSITE Generative artificial intelligence. (n.d.). In Wikipedia. [Article Link]
