The EU AI Act (Regulation (EU) 2024/1689) is the European Union's binding legal framework for artificial intelligence. It's the first comprehensive AI law of its kind anywhere, and for anyone producing AI-generated fashion or product imagery for the European market, one part of it matters more than the rest: Article 50.
Article 50 covers disclosure, and it actually splits into two obligations. Providers of AI systems that generate synthetic image, audio, video, or text content have to mark the output as machine-readable and detectable as artificially generated, a technical marking, not necessarily something a viewer sees directly. Deployers showing content that qualifies as a "deep fake," meaning it resembles a real person, object, place, or event closely enough that it would falsely appear authentic to a viewer, carry a separate, stricter duty: a clear, human-visible disclosure that the content is artificially generated or manipulated. For fashion brands using AI models, AI-generated campaign imagery, or AI-retouched product photography realistic enough to pass as an ordinary photo, that second duty is the one that matters: visible, accessible disclosure that the image isn't a traditional photograph.
The compliance deadline is August 2, 2026. After that date, non-compliance can carry penalties of up to €15 million or 3% of a company's global annual turnover, whichever is higher (Article 99(4)(g)), the Act's second-highest penalty tier, one step below the €35 million / 7% tier reserved for banned "unacceptable risk" practices like social scoring, but still a serious, binding enforcement level, not a formality.
Where Article 50 sits in the Act's structure
The AI Act's core structure prohibits a short list of "unacceptable risk" practices outright (Article 5, things like social scoring) and imposes heavy, system-level requirements on "high-risk" AI systems (Chapter III, think hiring or credit-scoring tools). Article 50's disclosure duties are a separate, narrower layer: they attach to specific outputs and interactions, chatbots, emotion recognition, deep fakes, AI-generated content on matters of public interest, rather than to a system's overall risk classification, and they apply on top of, not instead of, any high-risk obligations that also apply. AI-generated imagery falls under this disclosure layer, which the European Commission's own materials now group under "transparency risk" (still often called by the older shorthand "limited risk"): the obligation is disclosure rather than pre-approval or a ban. That's a lighter compliance burden than what applies to high-risk AI systems, but it's still a binding legal requirement with real penalties attached, not a voluntary best practice.
What counts as acceptable disclosure
The Act doesn't prescribe one exact label, but Article 50(5) does specify what disclosure has to achieve: it must be provided in a clear and distinguishable manner, at the latest at the time of first interaction or exposure. In practice, that rules out burying disclosure in a footer link three clicks away. There's a narrower leeway in the Act for content that's part of an "evidently artistic, creative, satirical, fictional" work, disclosure there just has to not hamper how the work is displayed or enjoyed, but ordinary commercial fashion and product photography won't qualify for that exemption. For that content, the safer reading is that disclosure needs to be genuinely clear and noticeable, not minimized to whatever fits without disrupting the shot. The workable approach is a visible, legible label near the image, consistent across a brand's channels.
Why this isn't optional for AI fashion photography specifically
The Act's recitals and subsequent guidance treat photorealistic AI-generated people as a specific area of concern, close enough to deepfake territory that "it's obviously AI, everyone can tell" isn't a defense. As AI-generated fashion models have gotten indistinguishable from real photography, the law has moved to require disclosure precisely because the images no longer disclose themselves. This is also where Article 50 overlaps with the broader digital twin conversation: an opt-in, brand-authorized AI model reference and an unauthorized AI likeness of a real person can produce visually identical output, and disclosure is the mechanism that applies regardless of which one generated the image.
Who it applies to
The Act applies to any provider or deployer placing AI systems or AI-generated content in front of users in the EU, regardless of where the company is headquartered. A US or UK brand selling into German or French wholesale accounts is in scope the same as an EU-based one. The obligation most naturally sits with whoever generated or manipulated the image; how far it extends to marketplaces and retailers simply republishing that image is a genuinely unsettled question, not something to assume away. The safer default is for the brand producing the imagery to handle disclosure at the source, rather than leaving it to whichever channel happens to display it last.
At Brandmachine, every export carries the information needed to disclose correctly, and Product Studio's outputs are built to make labeling straightforward rather than an afterthought bolted on at the end of a campaign. Compliance is a design constraint we build for, not a checkbox we hope customers remember on their own.