Article 50 of the European Union's Artificial Intelligence Act (AI Act) took effect on August 2, 2026. This is an EU regulation, not global law - it applies directly within EU member states. For a business operating entirely outside the EU, it doesn't automatically apply.
If you work with Claude, ChatGPT, Artlist, or similar tools for EU clients, or your site/store offers goods or services to people in the EU, this still applies to you - regardless of where you're registered yourself. What matters is where the output gets used and who it's offered to, not where you sit while working. Here's what that actually means for everyday work, with no legal jargon.
Provider or deployer - this is where most people get confused
This is the most common mix-up, so it's worth clearing up right away. The law distinguishes two completely different roles, and your obligation depends on which one you have.
The provider is the company that builds the AI tool itself - Anthropic (Claude), OpenAI (ChatGPT), Artlist. Their obligation is to technically mark the tool's output - embedding an invisible, machine-readable "watermark" into generated text, images, audio, or video, so it can later be determined the content was artificially generated.
The deployer is you - the designer, agency, or store owner using those tools at work. Your obligation is far narrower than most people assume - it doesn't apply to every piece of AI content you create, only to specific situations, described below.
In other words: when Claude writes a product description for your store, the technical marking of that output is Anthropic's obligation, not yours. Your obligation only kicks in once you use that content in one of a few specific situations.
What you actually need to label
Here's a quick rundown of the most common situations in design and agency work, and whether your labeling obligation applies.
| Type of content | Your obligation as a deployer |
|---|---|
| Text from AI (product description, landing page) | No, unless it's published as information on matters of public interest with no human editorial review |
| Photo touched up with AI (retouching, background removal) | No - standard editing is exempt |
| Image generated from scratch (AI banner, illustration) | Marking is the tool's job; you usually don't need extra disclosure, unless it resembles a real person or event |
| Video or image resembling a real, recognizable person | Yes - that's a deepfake by legal definition and must be clearly disclosed |
| AI voice mimicking a real person | Yes, same rule as deepfake content |
A deepfake is legally defined as image, audio, or video content that AI generates or manipulates, resembling real people, places, or events, that could mislead someone into believing it's authentic. For content that's part of an evidently artistic, creative, satirical, or fictional work, a minimal, non-intrusive disclosure is enough (say, an icon shown for a few seconds in a video).
When AI just "helps", labeling isn't required
The law explicitly exempts situations where AI performs an "assistive function for standard editing" or doesn't substantially alter the input data you provided.
A practical example from design work: retouching a product photo - brightening, removing a blemish, sharpening - counts as standard editing and falls outside this obligation. Fully replacing the background or generating a completely new photo from text goes beyond that - that's where the technical marking obligation kicks in, but it still sits with the tool generating the content, not directly with you.
If you have an AI chatbot or assistant on the site
This matters especially if you build or use an AI chatbot for customer support, bookings, or answering inquiries - the first part of Article 50 (paragraph 1) covers exactly that. A system that directly "talks" to a visitor has to be designed so that person knows they're talking to AI, unless that's already obvious from context.
In practice, this usually just means a simple, visible indicator - say, "AI Assistant" in the header of the chat window, or a short line at the start of the conversation ("This is an automated assistant, here to answer your questions"). There's no need for complicated legal text - just a clear, noticeable label before the conversation actually starts.
What if you only work for the local, non-EU market
If both you and your client operate entirely outside the EU - say, you're building a site for a local business that only sells to domestic customers - Article 50 obligations technically don't apply to you. This is an EU regulation and carries no legal force outside the EU market, unless you choose to follow it yourself as good practice.
The line starts to matter the moment any EU element shows up - a site that also sells and ships goods to customers in the EU, a service offered to EU citizens too, or a business with a branch or office in an EU member state. In that case, that same site or product suddenly falls under this law for that part of the business, even if the owner is based outside the EU. When in doubt, it's simpler to apply the same standard everywhere than to try to draw a line by market.
The European Commission has announced a targeted transition period, with a potential date of December 2, 2026, for part of the technical marking and detection obligations covering AI systems that were already in use before August 2, 2026. This mainly concerns the tool providers themselves, not you as a user - but it's worth knowing this exists, in case someone mentions a "transition period" in the context of the AI Act.
One sentence that covers the gray area
Even when the law doesn't strictly require disclosure, a short note in your site's footer or legal page removes ambiguity and builds trust with visitors, with no extra work involved:
"Some of the content on this website was created or processed with the help of artificial intelligence tools."
This isn't always legally required in your specific case, but it's a simple, one-time move that puts you on the safe side, instead of guessing whether a given situation falls into the gray zone.
Fines - briefly
Failing to meet Article 50 obligations can trigger a fine of up to €15 million or 3% of the company's total worldwide annual turnover, whichever is higher. For small and medium businesses, including startups, the law applies the lower of those two amounts, not the higher one - which significantly lowers the real risk for smaller agencies and stores.
These are ceilings set for the most serious, deliberate violations, not automatic penalties for every slip-up - enforcement practice is still taking shape, and oversight sits with each member state's national authority.
If nothing else, add one sentence to your site's footer today about using AI tools in your work.
