EU AI Act 2026, explained in plain English
Here's the one-sentence version: from 2 August 2026, if your website or store shows AI-generated or AI-altered content that could pass as real, you need a clear, visible label the first time someone sees it. No hidden metadata, no fine print — a label a normal person actually notices.
Everything else in the EU AI Act is either background (why this rule exists) or detail (exactly how to apply it to your situation). This guide covers both, in plain language, with no legal degree required.
Not legal advice — this is a practical explainer. Confirm your specific obligations with counsel.
What is the EU AI Act, actually?
The EU AI Act (Regulation (EU) 2024/1689) is the European Union's law on artificial intelligence. It's broad — it covers everything from banned AI practices to rules for "high-risk" AI systems used in hiring or credit scoring.
But if you run a website, an online store, or a marketing team, the part that affects you day-to-day is Article 50 — the transparency chapter. It's the rule about telling people when they're looking at AI-made or AI-changed content, so they aren't misled into thinking a synthetic photo, video, or voice is real.
Think of it like a nutrition label, but for content: it doesn't ban anything, it just requires you to say what's in it.
Who does this apply to? (Provider vs Deployer)
The law splits responsibility by role, not by company size:
| Role | Who this usually is | What they must do |
|---|---|---|
| Provider | The AI tool that generates the content (Midjourney, ChatGPT, Adobe Firefly, etc.) | Mark outputs in a machine-readable way at creation (Art. 50(2)) — watermarks, C2PA-style signals |
| Deployer | The business that publishes the AI content to its audience — a store, an agency, a brand | Show a clear, visible label at first exposure (Art. 50(4) / 50(5)) |
| End user | Your shopper, reader, or follower | Must be able to see the disclosure without hovering, clicking, or scrolling |
This is the part people miss: if you use an AI tool to make an image and then publish it under your own brand, you are the deployer — not the AI company. Midjourney's watermark (if any) does not cover you. The responsibility to disclose sits with whoever puts the content in front of an audience.
If you run client work or manage multiple brands, the deployer role can shift per property. See our agency-focused breakdown →
If you run a store, see our merchant compliance checklist →
The key date: 2 August 2026
Deployer transparency duties under Article 50 apply from 2 August 2026. There's a separate grace period for some provider-side machine-readable marking (until 2 December 2026), but that does not push back your deployer labeling duty — those are two different clocks.
Content published before 2 August 2026 generally doesn't need retroactive labeling, though the European Commission encourages it where practical.
What actually counts as "AI content" here?
The law focuses on content that could reasonably be mistaken for something real — often called a deepfake in plain terms, even outside the classic "fake video of a politician" scenario.
Usually needs a label:
- A fully AI-generated product photo or lifestyle scene
- A real photo with the background swapped, an object added/removed, or a synthetic person placed into it
- Realistic AI imagery in ads, landing pages, or case studies
Usually doesn't need a label:
- Minor retouching, color correction, cropping, or exposure tweaks
- Obviously stylized illustrations, abstract graphics, or decorative AI art
Rule of thumb: if a viewer could reasonably think "this is what it really looked like," label it. If it's clearly illustrative, it's usually not the priority — ask counsel on the borderline cases.
What you actually have to do
Three things, in order:
- Find it. Identify which images (or video/audio) on your site are AI-generated or materially AI-altered.
- Say it clearly. Add a short, plain-language label — "AI-generated image" or "AI-altered photo" — not just an icon, and not buried behind a hover state.
- Show it early. The label must be visible at first exposure — the moment someone sees the content, not after they click into a details tab or scroll to a footer.
What doesn't count
A common misconception: "I embedded IPTC / C2PA metadata in the file, so I'm covered." Metadata is mostly a provider obligation, and on most platforms (including Shopify's CDN), that metadata gets stripped on upload anyway. A shopper never sees a file's hidden metadata — they see what's rendered on the page. That's why the rule requires a visible label, not an invisible tag.
Why this matters: the fines
Enforcement sits with national authorities under Article 99 of the AI Act. Transparency obligations (like Article 50) generally fall under the "other Act obligations" tier:
| Tier | Maximum fine |
|---|---|
| Prohibited AI practices | Up to €35M or 7% of global turnover |
| Other Act obligations (transparency, Art. 50, often discussed here) | Up to €15M or 3% of global turnover |
| Misleading information to authorities | Up to €7.5M or 1% |
Exact application depends on the facts and the authority involved — this isn't a per-image fine calculator. But "we're a small store" isn't a built-in exemption, and beyond fines there's reputational and platform risk to consider.
Where to go next, depending on who you are
| You are… | Read this |
|---|---|
| An agency managing client sites or campaigns | EU AI Act 2026: what agencies need to know → |
| A merchant or brand running a store | EU AI Act 2026: how to be compliant as a merchant → |
| Looking for the full toolkit across every channel | aiact.solutions |
The toolkit, at a glance
Different content lives in different places, so the labeling approach changes by channel:
| Channel | Label type | Best tool |
|---|---|---|
| Shopify / WordPress product photos | Visible storefront overlay | EU AI Label |
| Any website, scanned automatically | Site-wide badge script | AI Act Icon |
| Email, social, ads, PDFs, decks | Burned-in label in the creative | EU AI Icon |
| Not sure where to start | Full overview of all three | aiact.solutions |
FAQ
Is the EU AI Act only about deepfakes and politics?
No. Article 50's transparency duty covers everyday AI content too — product photos, marketing images, and ads — anywhere realistic AI content could mislead someone about what they're looking at.
Does this apply to me if my business isn't based in the EU?
Often yes in practice, if you sell to or reach people in the EU. Confirm your specific exposure with counsel.
Does an AI tool's built-in watermark cover my obligations?
No. That's a provider-side signal, and it doesn't create a visible, human-readable label on your own storefront or marketing. You still need your own clear disclosure.
What's the fastest way to get compliant-looking labels live?
For Shopify/WordPress product photos, EU AI Label adds visible overlays without editing theme code. For one-off images (social, PDFs), EU AI Icon burns a label directly into the file. See the Shopify AI Act checklist for a step-by-step.
Where can I read the actual law?
Article 50 text is available via the AI Act explorer, and the European Commission publishes an official FAQ.
Get started
Explore the full compliance toolkit at aiact.solutions, or jump straight to the tool that matches your channel: EU AI Label for stores, EU AI Icon for one-off files, and AI Act Icon for whole-site coverage.
This article supports transparency workflows under Regulation (EU) 2024/1689. It is not legal advice and does not guarantee compliance. Confirm obligations with counsel for your specific case.