TL;DR
- EU: disclose AI-generated or manipulated deepfake-style image/video/audio content where it could appear authentic.
- U.S.: avoid misleading consumers under FTC principles.
- New York: disclose synthetic performers in ads starting from June 9, 2026; get consent for model digital replicas; avoid unauthorized commercial use of deceased individuals' likenesses.
With the EU AI Act introducing some of the world's first broad rules around AI transparency, brands are beginning to ask new questions about how AI-generated images and videos should be disclosed. What actually needs to be labeled? When does the law apply? Who does the law apply to? And what does it mean for e-commerce teams creating visual content?
Let's be honest, though, this is an overwhelming topic to dive into. I don't have a legal background, and while I've had the opportunity to learn from experts like Kelsey Farish during her sessions at FLOW Europe, I'm far from one myself. Nothing in this article should be taken as legal advice. Think of it as a research-based starting point reflecting where things stand as of late July 2026.
That being said, I did spend a lot of time reading through the legislation, guidance, and commentary to better understand how the EU AI Act (and similar developments in the U.S.) are shaping AI-generated content in real-time.
For many brands, the question may soon no longer be whether to label AI-generated content. Based on the direction of today's legislation, particularly in the EU and New York in the U.S., it increasingly looks like some level of transparency will become part of doing business.
The more interesting questions are the ones that come after that.
- How do you label content in a way that reflects your brand?
- How visible should those disclosures be?
- Can you be transparent while still preserving the creative integrity of the image?
Those are the conversations I think we'll see more of over the next few years. The regulations will almost certainly continue to evolve, but so will the way brands communicate AI use to their customers. Finding an approach that is compliant, consistent, and feels authentic to your brand may end up being just as important as the label itself.
The sections below are the key takeaways from my research, along with some of my own interpretations. Think of them as a starting point rather than the final word. Every organization will still need to interpret the requirements in the context of its own products, workflows, and legal obligations.
EU: AI Act
For e-commerce brands, the most relevant EU rule is the AI Act's transparency obligation for deepfakes. If a brand uses an AI system to generate or manipulate image, audio, or video content that could falsely appear authentic, the deployer must disclose that the content was artificially generated or manipulated. The AI Act defines "deep fake" as AI-generated or manipulated image, audio, or video content resembling existing persons, objects, places, entities, or events that would falsely appear authentic or truthful.
That definition initially sent me down a bit of a rabbit hole. Why not just say "AI-generated content"? It turns out that's intentional. The legislation isn't trying to regulate every use of AI. Instead, it's focused on AI-generated or AI-manipulated media that could reasonably be mistaken for something real. In other words, the emphasis is less on how the content was created and more on whether it could be perceived as an authentic depiction of a person, object, place, or event. The legal definition is therefore much broader than the way most of us use the word "deepfake" in everyday conversation.
What isn't exactly entirely clear is what "manipulate" means in this context, most likely referring to generating a fake model that doesn't exist, making a person appear to say/do something they didn't, creating fake events/scenes, materially altering products or environments in deceptive ways, or synthetic influencer/deepfake-style content.
For product imagery and campaign visuals, this means disclosure most likely is required where AI-generated or AI-manipulated visuals present people, products, places, or events in a way consumers could reasonably read as real. The European Commission also notes that deepfakes and certain AI-generated public-interest content should be clearly and visibly labeled, with the transparency rules coming into effect in August 2026.
Who does the AI Act apply to?
By now you're probably wondering: Does this actually apply to me?
The first thing to know is that the AI Act doesn't treat everyone using AI the same. Instead, it divides responsibility between two groups: AI providers and AI deployers.
AI providers are the companies building the technology itself. Think OpenAI, Adobe Firefly, Google Gemini, Midjourney, and the other companies developing generative AI models. Under the AI Act, they're generally responsible for making sure AI-generated outputs are machine-readable where technically feasible. In practice, that doesn't necessarily mean a visible watermark. Instead, it refers to technical markers such as metadata or content credentials that allow software and other systems to recognize that AI was involved.
AI deployers are the businesses actually using those AI tools to create and publish content. If you're a fashion brand creating AI campaign imagery, an agency producing AI advertisements, or a company like Pixelz helping generate e-commerce visuals, you'd likely fall into this category.
For deployers, the focus shifts from technical markers to public transparency. If AI-generated or AI-manipulated content could reasonably be mistaken for an authentic depiction of a person, product, place, or event, the AI Act generally requires a clear disclosure that the content was artificially generated or manipulated.
There is one notable exception. Where content is clearly artistic, fictional, satirical, or similarly creative, the disclosure may be presented in a way that doesn't interfere with the viewing experience. For standard e-commerce advertising, however, that exception is likely to be much narrower because the purpose is commercial product promotion.
What if my company isn't based in Europe?
The AI Act isn't just for European companies. Much like the GDPR, it's less about where your headquarters are and more about where your content is being shown. If you're a U.S. retailer, agency, or platform selling products to customers in the EU or publishing AI-generated commercial content that reaches the European market, these transparency obligations may still apply.
So the real question isn't, "Where is my company based?" It's, "Is my AI-generated content reaching people in the EU?" If the answer is yes, it's worth understanding what your responsibilities are before the rules take effect on August 2, 2026.
Who will enforce it?
The AI Act doesn't create an "AI police" monitoring every corner of the internet. Instead, enforcement falls to national market surveillance authorities decided by each EU Member State.
But that naturally raises a few questions. Who are these authorities? Do they have the resources to keep up with the sheer volume of AI-generated content being published every day?
In reality, enforcement is likely to be largely complaint-driven, at least initially. A consumer, competitor, or organization can report potential non-compliance, prompting an authority to investigate. What those investigations will look like and how actively different Member States choose to enforce the rules remains to be seen...
U.S.: More state-led than federal action
At the U.S. federal level, there is not yet one EU-style AI Act for e-commerce visuals. The most relevant federal baseline is still the FTC's truth-in-advertising principle: ads, endorsements, and testimonials must be truthful and not misleading. That matters if AI visuals create a false impression about a product, person, endorsement, review, or consumer experience.
New York: Synthetic performers in ads
New York's S.8420-A/A.8887-B is especially relevant for e-commerce campaigns. It requires advertisements to disclose the use of a synthetic performer and imposes civil penalties of $1,000 for a first violation and $5,000 for subsequent violations. The law applies to ads using AI-generated human-like performers and took effect June 9, 2026. A synthetic performer in this case is a human-like digital asset created by generative AI or a software algorithm that engages in audiovisual and visual performances.
This is important if a brand uses AI-generated "models," "people," influencers, or human-like figures in product ads shown to New York audiences. The law's focus is disclosure, not an outright ban. But what does the labeling mean here? It felt a bit clearer in the EU act to me… But apparently, anything watermarked. Some say a light gray in the corner is asking for trouble, but I'm not sure. And I couldn't find anything on the machine reading ability in this context.
One question I kept coming back to while reading the legislation was: what does that disclosure actually look like? The EU AI Act is a bit more specific about transparency obligations, but New York's law is less detailed about how brands should communicate that a performer is synthetic.
A visible disclosure or watermark is generally understood to satisfy the intent of the law, as I read it, but there is little guidance today on exactly how prominent that disclosure should be. Unlike the EU AI Act, I also couldn't find an equivalent requirement for machine-readable markers or embedded metadata. My guess is that we'll see more guidance and probably emerging industry conventions as brands begin putting these requirements into practice.
Who does this apply to?
So, who actually carries the responsibility? The law places that responsibility on the person or business producing or creating the commercial advertisement. So, whether you're a solo creator, an agency, or a global fashion brand, if you're knowingly using an AI-generated synthetic performer in an advertisement shown to a New York audience, you'll generally be responsible for including the required disclosure. It's not about where your company is headquartered; it's about where your advertisement is being seen.
The disclosure also needs to be conspicuous, not hidden in tiny text or tucked away where consumers are unlikely to notice it. A first violation can result in a $1,000 civil penalty, with subsequent violations increasing to $5,000.
For fashion and e-commerce brands, the question is becoming fairly simple: Does this ad feature a human-looking person who was generated by AI rather than photographed?
Who will enforce it?
Just like the EU AI Act, there's still an open question around enforcement. The law establishes disclosure requirements and civil penalties, but it doesn't create a dedicated AI regulator or explain exactly how violations will be identified in practice. Will enforcement be driven by complaints from consumers or competitors? Will regulators actively monitor advertising? We'll likely get those answers through the first enforcement cases, just as we will in Europe.
Still open to interpretation…
One final note and maybe the biggest gray area of them all. New York's law applies to advertisements, not necessarily every AI-generated image that appears on a website. A paid social campaign or promotional banner clearly fits that definition. But what about a standard product detail page?
If an AI-generated model is simply being used to showcase a product on a PDP, it's still uncertain whether regulators would consider that an advertisement under the law. For now, it's a bit of a waiting game, and we'll likely need additional guidance and more legal precedent before there's a definitive answer.
New York: Post-humous & digital replicas
New York's S.8391/A.8882 expands protections around deceased individuals' likenesses and digital replicas. It relates to the use of deceased performers' or personalities' digital replicas without authorization from the applicable right holder. For e-commerce, this matters if a brand wants to use an AI-generated version of a deceased celebrity, artist, model, or public figure in commercial visuals. This one probably falls more into the "good to know" category, but as AI makes it easier to recreate people digitally, it doesn't feel quite as far-fetched as it once did.
New York: Fashion Workers Act
New York's Fashion Workers Act is directly relevant to fashion and e-commerce brands working with models. The New York Department of Labor states that, as of June 19, 2025, prior power-of-attorney agreements covering a model's digital replica are no longer valid. Legal summaries of the Act explain that brands and model management companies need clear written approval before creating or using a model's digital replica, including details such as scope, purpose, pay, and duration.
This is one that's well worth keeping on your radar, particularly if you're a fashion or e-commerce brand operating in the U.S. AI-generated digital twins or digital replicas are becoming more common, and so are the legal expectations around consent.
So, where does that leave us?
That's where things stand today. These are the regulations that currently shape how AI-generated media is labeled, along with the other laws that may affect e-commerce brands using AI in their content.
Is everything perfectly clear? Not quite and not even close. The legal intricacies are still evolving, and many parts of these regulations remain open to interpretation. While the text of the law gives us somewhat of a framework, many of the practical questions — how disclosures should appear, what regulators will prioritize, and how enforcement will work in practice — will likely only become clearer over time through guidance and the first landmark enforcement cases.
At Pixelz, we've decided not to wait. We're rolling out labeling across our website, social channels, and other customer-facing content. Not simply because the law nudged us in that direction, although it certainly accelerated the conversation, but because we believe transparency is the right long-term approach.
As AI becomes a normal part of creative production, transparency will likely become just as normal. The regulations will continue to evolve, guidance will continue to emerge, and the first enforcement cases will shape how many of these rules are interpreted in practice. Until then, being open about how content is created feels like a habit worth building now rather than later.