Pennsylvania joins the states with AI ad label laws, but viewers want far more disclosure

Pennsylvania Governor Josh Shapiro signed an AI advertising disclosure law on October 22, making Pennsylvania the third US state with one after New York and California. In our X poll, 87.8% of 1,333 voters said films and ads should always say when they used AI. Two larger surveys this fall found 90% and 74% support for disclosure.

The laws themselves mostly require a label only when AI could mislead someone: a realistic synthetic person selling a product, or content that gives a false impression of it. So the audience bar sits well above the legal one, and a creator who clears only the law can still look like they hid something. What most coverage of these laws skips is the credit itself. A useful AI disclosure names the task AI did, what viewers actually see or hear, and the people who did the work.

What the poll and the surveys show

Our poll ran on X from October 21 to October 27 and asked: "Should films and ads have to say when they used AI?" A paid boost brought in most voters, and X polls carry no demographics, so 87.8% is a strong signal from the people we reached, not a measure of all audiences. One voter put the mood simply: "It's getting harder to tell what's real anymore." Only 3.8% picked the middle options, mostly AI or only faces and voices, and 8.5% chose "No, judge the work."

Proper surveys point the same way. The British Council asked 2,001 people aged 16 to 24 in the UK, and 90% said film and TV should be required to disclose AI use. Cint, working with Advertising Week, found 74% of about 1,000 US adults think brands and creators have a duty to disclose AI in ads, and 44% would trust a brand less if it did not. In the Cint data, AI-generated people were the most disliked use and realistic AI landscapes the most liked.

The opposing view has evidence too. In 16 experiments with 27,000 people, Michigan Ross researchers found that an AI label cut ratings of identical creative writing by 6.2% on average. Disclosure has a cost, which is why the credit matters.

What the law actually asks

New York's synthetic performer law has been in force since June 9. It covers ads featuring a digitally made person who is not a real, identifiable one, requires a conspicuous disclosure and sets fines of $1,000 for a first violation and $5,000 after that. California's SB 1050, signed on September 16 and sponsored by SAG-AFTRA, takes effect on January 1. It applies when a synthetic performer appears prominently, demonstrating the product, voicing the narration or reacting to it, and the label must say something substantially similar to "this performance features a synthetic performer" or "no human performer is depicted." Ads for films and games get an exemption when the synthetic performer is used the same way as in the work.

Pennsylvania's law, SB 806, takes effect on December 21 and reaches further than either. It covers any synthetic content, including images, audio and text significantly modified or generated by AI, but only when it creates a false impression of the product. The disclosure has to say the ad contains AI-generated or AI-modified content, appear when that content is first shown and use the same medium.

The EU AI Act's labeling duty, which applied from August 2, targets deepfakes and allows a lighter touch for evidently artistic work. The UK's Advertising Standards Authority says there is no blanket duty to disclose AI in ads. YouTube, TikTok and Meta ask for labels on realistic people, events or scenes. Royal Mail's Christmas 2025 ad used an AI narrator without saying so, and when that surfaced in October the row turned on whether it copied voice actor Paul Vaughan, which the AI voice companies deny. No rule required a label, and the backlash came anyway.

Grimes credited the people, not the task

Fans asked whether Grimes's "Accelerando" video, released on October 9, was AI. Her spokesperson told Business Insider the song and vocals used no AI and that the video contains "AI-generated elements" that were "selectively incorporated as a supplementary tool within this much larger, artist-driven VFX pipeline, rather than serving as a replacement for the creative or technical work." According to the spokesperson, more than 70 artists and production professionals worked on it over two to three months, and the end credits list at least three AI artists alongside the dancers, choreographer and hair and makeup team.

That is more than most productions say, and it still left the question open: asked what the AI artists contributed, the spokesperson defended their work without naming shots or tools. The credit named the people but not the task.

The formats already on the table

Voluntary standards are already being written. The Archival Producers Alliance suggests treating AI material like third-party footage in the end credits: name the tool and version, the source it drew on, what was altered and the people responsible, with roles such as synthetic voice engineer. AIMICI's proposed standard for UK film and TV starts with a three-part statement for audiences covering where AI was not used, where it was used and who was accountable. Responses are due on November 19, with interim findings promised for November 30.

Human Provenance in Film sorts each release into No AI Used, Assistive AI or Generative AI, and any generative use puts a film in the top class. On November 2 it adopted optional descriptors such as generated voice and digital replica, so a declaration can separate generated scenery from a generated performance. Its rule is worth borrowing: missing facts never establish No AI Used.

What a useful AI credit says

Our proposal, not a standard, is three lines. First, what AI made or changed, by task and shot count. Second, what stayed human, especially performances, voices and music. Third, who did the AI work and who signed it off. An illustrative end-credit block, used by no real production:

AI disclosure: generative AI made the background landscapes in six shots and the crowd extensions in two. AI changed one actor's age in the flashback, with the actor's written consent. Every performance, voice, the score and the choreography are human. AI artists: [name], environments, and [name], compositing and cleanup, supervised by [VFX supervisor].

For ads and social posts the short version goes on screen, next to the element: "AI-generated voice" or "AI-generated background." California will require that kind of placement for synthetic performers anyway.

What it means for AI creators

• The legal minimum is not the audience minimum. A video with no label, or a vague one, now reads as something to be suspicious of, even when no law required anything.
• Naming the task protects the work. Viewers dislike AI people far more than AI scenery, so a credit that says the landscapes were generated and the performances were not tells them what they actually want to know.
• Naming the people is the part of disclosure that works in your favor. It turns an AI credit from a confession into a crew list.

Our take

Disclosure is turning into a craft credit, and the creators who write a specific one first will set the norm instead of having a vague label written for them by a regulator or a platform.

Timeline

• Date: Tue 9 Jun 2026. Event: New York's synthetic performer ad law takes effect; Source: NY Governor's office
• Date: Sun 2 Aug 2026. Event: EU AI Act labeling duties for deepfakes apply; Source: European Commission
• Date: Wed 16 Sep 2026. Event: California SB 1050 signed; Source: Governor of California; leginfo
• Date: Fri 9 Oct 2026. Event: Grimes releases the Accelerando video; Royal Mail AI voice reported; Source: Business Insider; Deadline
• Date: Tue 13 Oct 2026. Event: British Council survey: 90% of UK 16 to 24s back required disclosure; Source: British Council
• Date: Tue 20 Oct 2026. Event: Cint and Advertising Week survey: 74% of US adults; Source: Cint
• Date: Wed 21 Oct 2026. Event: Our X poll opens (paid promotion); Source: X
• Date: Thu 22 Oct 2026. Event: Pennsylvania SB 806 signed; Source: Pennsylvania General Assembly
• Date: Tue 27 Oct 2026. Event: Poll closes at 1,333 votes, 87.8% Yes, always; Source: X
• Date: Mon 2 Nov 2026. Event: Human Provenance in Film adopts optional descriptors; Source: humanprovenance.film
• Date: Thu 19 Nov 2026. Event: AIMICI consultation responses due; Source: aimici.co.uk
• Date: Mon 21 Dec 2026. Event: Pennsylvania law takes effect; Source: Inside Privacy; National Law Review
• Date: Fri 1 Jan 2027. Event: California SB 1050 takes effect; Source: leginfo

Sources and what is unconfirmed: our Generated Voices W1 poll on X (read at close); British Council release, 13 Oct; Cint newsroom, 20 Oct; Berg and colleagues, Michigan Ross, Jan 2026; Business Insider, 13 Oct; Deadline and The Independent, 9 Oct; New York Governor's release, 9 Jun; California SB 1050 bill text and Governor's release, 16 Sep; Coblentz and Morgan Lewis client notes on SB 1050; Pennsylvania General Assembly Act 99 page; Inside Privacy, 9 Nov, and National Law Review, 2 Nov, on SB 806; European Commission AI Act pages; ASA guidance; Archival Producers Alliance credit suggestions; aimici.co.uk; humanprovenance.film update of 2 Nov. Unconfirmed: which Accelerando shots used AI (not disclosed); the Pennsylvania statute text (summarized from two law firm notes, the official text page was unavailable); how New York will read "conspicuous". This is our reading of the rules, not legal advice. We did not contact Grimes's team, Royal Mail or the standards bodies.

Your Voice

Quoted from public posts on X and credited by handle. If one of these is yours and you want it removed, message @the__generated on X and it comes down.

Got something to say? Tag @the__generated on X