Pennsylvania will require labels on advertising made with artificial intelligence, The Verge reports — adding another US state to the list of jurisdictions that have concluded disclosure, rather than prohibition, is the practical answer to synthetic persuasion.
The vehicle is a deceptive advertising act, and its core obligation is easy to state: an ad that was generated or materially altered by AI has to say so, in a place a viewer will actually see it.
That choice is not an accident of drafting. Outright bans on synthetic political speech have run into First Amendment challenges in American courts, and legislatures have responded by retreating to transparency duties, which survive legal review more comfortably because they leave the speech itself intact and only attach a condition to it. Europe took a comparable route in the AI Act, which imposes transparency obligations on synthetic media rather than forbidding it.
Whether a label does any work, though, depends on mechanics a headline cannot settle. Three questions decide the outcome.
What counts as AI-generated? A fully synthetic video is the easy case. A cloned voice, a photograph retouched with generative fill, or a script drafted with a chatbot and performed by a human are all harder — and every change to the definition moves the compliance burden.
Who carries the duty? The advertiser who commissioned the spot, the agency that produced it, or the platform that sold the impression. If platforms are liable, they will demand declarations at upload. If only advertisers are, enforcement depends on somebody auditing finished creative after the fact.
What is the consequence of a missing label? A fine that is small relative to a campaign budget is a rounding error. A takedown requirement is a real incentive. Nothing enforceable is a press release.
The technical layer is the weakest link in every labelling regime built so far. Provenance standards such as C2PA's Content Credentials can travel with a file, but they are stripped by a re-encode, a screenshot or a crop. Detection tools are probabilistic and inconsistent across generators. In practice, most disclosed AI advertising will be disclosed because the advertiser chose to comply — which means the law's bite lands on the actors least likely to care.
There is also the patchwork problem. State-by-state rules with different definitions, different scopes and different penalties push national advertisers toward the strictest common denominator, or toward simply not running synthetic creative in states that demand paperwork. Federal legislation would flatten that, and has not arrived.
None of this makes the Pennsylvania law pointless. A disclosure requirement sets a norm, gives regulators a hook, and makes the honest case cheap while making the dishonest case expensive to defend in public. But a label is a floor, not a solution: it tells a viewer that a machine made the thing they are looking at, and nothing at all about whether the claim inside it is true.
The test is narrow and concrete. In the first election cycle this act covers, can a voter watching a fifteen-second spot on a phone actually see the disclosure — or will it live in a terms page nobody opens?
