N Y Targets AI Models in Product Ads

A recent change to a New York State advertising law may be well-meaning but could nonetheless limit generative AI innovation and impose regulatory burdens on retailers.

New York State Sen. Michael Gianaris (D) and Assemblywoman Linda Rosenthal (D) sponsored the measure, which amended the state’s advertising rules to require clear notices when ads contain AI-generated images of people.

The law will make AI-generated photography far less effective, since the prescribed disclosure is, in effect, a warning that invites shoppers to distrust an otherwise lawful image.

An AI-generated image of a model wearing shoes could violate the New York law.

Amazon Responds

The new law may already be affecting ecommerce.

Amazon told third-party sellers to identify product content featuring AI-generated images of people before uploading it, according to CNBC.

The marketplace may then display a notice to shoppers, although Amazon has not explained exactly when the label will appear or how prominent it will be.

Amazon’s response demonstrates how one state can influence ecommerce practices nationally.

A seller may not know where every shopper resides. And Amazon cannot easily determine from finished imagery whether a person was photographed, AI-generated, or substantially modified. Requiring sellers to supply that information enables Amazon to manage the risk.

The burden, however, falls on merchants. They must identify “synthetic” people, review existing creative, maintain production records, update metadata, and accept a warning that could impair the image’s effectiveness.

Related Rules

New York’s law appears unusual for commercial advertising. But the regulation is not entirely unprecedented in the context of AI generation.

California’s “AI Transparency Act” imposes origin and detection obligations on certain providers of generative AI systems.
Tennessee’s Ensuring Likeness, Voice, and Image Security (ELVIS) Act protects individuals from unauthorized AI replicas of their voices and likenesses.
Utah requires disclosure in certain consumer interactions involving generative AI.

Several states also regulate AI-generated political communications, while other laws target fabricated testimonials, intimate images, or unauthorized replicas of identifiable people.

Those rules address different risks, but collectively create a state-by-state compliance patchwork. Merchants selling nationally must interpret, track, and implement those requirements.

Product or Ad?

The N.Y. rule also leaves an open question: Is every product image an advertisement?

A product detail page is not the same as a television commercial, social media ad, or sponsored placement. Its images may show an item’s appearance, dimensions, use, or fit.

An AI-generated model wearing a shirt could serve the same merchandising function as a mannequin, an illustration, or a conventional catalog model. The model need not endorse or recommend the shirt, claim to have purchased it, or express any opinion about it.

Nonetheless, the legal risk is real.

The N.Y. statute applies broadly to advertisements concerning the use of people in products for sale.

Lifestyle photographs, on-model images, product demonstrations, and similar content help persuade shoppers. A regulator could therefore characterize them as advertising even when merchants regard them as information.

Amazon’s reported policy reinforces that possibility by applying the disclosure to product listings containing images of people.

Uneven Treatment

The New York regulation is also uneven.

A large retailer can hire a model, a photographer, a stylist, a studio crew, a retoucher, an agency, and an attorney. Any image resulting from that expensive photoshoot can be extensively manipulated through compositing, reshaping, color changes, artificial lighting, digital backgrounds, and other visual effects.

So long as the production remains sufficiently tied to a real performer, it may avoid the stigma imposed on a fully AI-generated model.

A smaller retailer can create a similar product image using generative AI from a single product photograph. But its image may require a warning, which will almost certainly negate its value.

Thus expensive human-based production receives ordinary treatment, while an affordable AI process receives scrutiny.

The difference has little to do with whether shoppers are being misled about the product. It reflects the method used to produce the image and the relative cost of that production.

Innovation

Generative AI removes one of ecommerce’s oldest competitive barriers.

For nearly three decades, polished product photography required a significant investment. Large retailers could produce extensive lifestyle and model imagery. Small merchants often had to settle for a few basic photographs. It was a significant advantage for enterprise businesses.

Generative AI changed that equation. With an accurate product photograph, a smaller merchant can now create model images, lifestyle settings, seasonal campaigns, localized variations, social media content, and product-in-use scenes in minutes and for pennies.

Generative AI is a form of creative democratization that levels the playing field between small businesses and large corporations by reducing the need for expensive photoshoots. New York’s law threatens that innovation.