The EU AI Act’s transparency rules, in plain language.

From 2 August 2026, Article 50 says you must tell people when they’re dealing with AI. Find out in two minutes whether it applies to you — and walk away with the exact wording to use.

days
hours
mins
secs

…until Article 50 applies. The rules were confirmed in June 2026 — the Digital Omnibus did not push them back.

Two-minute check

Does Article 50 apply to you?

Answer five quick questions about how your business uses AI. Nothing is sent anywhere — this all runs in your browser.

1. Do you run a chatbot, virtual assistant, voice bot or AI agent that talks directly to your customers or visitors?

Think website chat widgets, WhatsApp bots, AI phone answering, “AI concierge” tools.

2. Do you supply an AI tool or feature that generates text, images, audio or video for other people to use?

This is about being a provider — e.g. you’ve built and released a custom GPT, an app, or an AI feature others rely on. (Just using ChatGPT yourself doesn’t count here.)

3. Do you publish AI-generated or AI-edited images, audio or video showing real people, places or events in a way that could look genuine?

Deepfake-style marketing, AI voice-overs of real voices, “de-aged” or face-swapped footage, realistic AI images of real places.

4. Do you publish AI-written text to inform the public on matters of public interest?

News, current affairs, health, politics, consumer information. Not applicable if a person reviews and takes editorial responsibility for it.

5. Do you use AI to recognise people’s emotions or sort them by biometric characteristics?

Face-based age or sentiment analysis, “mood” detection, biometric audience categorisation.

The rules, briefly

What Article 50 actually says

Article 50 of the EU AI Act (Regulation 2024/1689) adds four transparency duties. They apply to any business whose AI use fits — not just “high-risk” systems, and open-source isn’t exempt. If you sell to, or have users in, the EU, it reaches you.

2 Aug 2026The day the transparency duties apply. Confirmed in June 2026 — not postponed.
4Situations covered: AI you chat with, AI-generated content, deepfakes / public-interest text, and emotion / biometric AI.
€15M / 3%Maximum fine for breaching Article 50 — €15 million or 3% of worldwide annual turnover, whichever is higher.
2 Dec 2026The one bit of breathing room: generative systems already on the market get until December for machine-readable marking.

Talking to people Providers · Art 50(1)

If your AI interacts directly with people (chatbots, voice assistants, agents), you must make sure they’re told it’s AI — clearly, and at the latest by the first interaction. Exception: where it’s genuinely obvious, or certain law-enforcement uses.

AI-generated content Providers · Art 50(2)

If you supply AI that generates text, images, audio or video, its outputs must be marked in a machine-readable way and detectable as AI-generated. Standard editing helpers that don’t substantially change the content are carved out.

Deepfakes & public-interest text Deployers · Art 50(4)

If you publish AI-made or AI-edited media of real people or events, or AI-written text informing the public on matters of public interest, you must disclose it. Lighter touch for clearly artistic or satirical work; not needed for public-interest text a human has reviewed and taken editorial responsibility for.

Emotion & biometric AI Deployers · Art 50(3)

If you use AI to recognise emotions or categorise people biometrically, you must inform the people exposed to it — and comply with data-protection law (GDPR). Note: emotion recognition in workplaces and schools is separately banned under Article 5.

Read the official text — European Commission: Article 50 · Article 5 · Article 99 (penalties). Plain-English guide: Article 50 explained (independent source).

Straight answers

Common questions

I only use ChatGPT / AI tools — am I caught?
Often less than you’d fear. Simply using AI internally isn’t the trigger. You’re caught mainly when your AI talks to your customers (a chatbot), or when you publish AI-generated content externally — deepfake-style media, or public-interest text without human editorial sign-off. The checker above sorts this out for you.
What’s the difference between a “provider” and a “deployer”?
Roughly: a provider makes or supplies the AI system; a deployer uses one under their own authority. Most small businesses are deployers. The chatbot and content-marking duties fall on providers; the deepfake, public-interest-text and emotion/biometric duties fall on deployers. You can be both.
Didn’t the “Digital Omnibus” delay all this?
No. The Digital Omnibus was adopted in late June 2026 and pushed back some high-risk deadlines — but Article 50’s transparency duties still apply from 2 August 2026. The only concession: generative systems already on the market before that date get until 2 December 2026 to meet the machine-readable marking requirement.
What does a compliant disclosure look like?
Clear, distinguishable, and shown at the latest at the first interaction or exposure — and it must meet accessibility requirements. What doesn’t pass: tiny footer text, a faint watermark, a label that flashes for an instant, or anything buried in the terms and conditions. The wording in your pack is written to be used up-front and in plain sight.
Is this legal advice?
No — and that’s important. This is a plain-language explainer and a set of templates to get you moving. Article 50 has genuine grey areas (what counts as “obvious”, “public interest”, “substantial alteration”), and your exact situation may need a professional view. Use this to understand the shape of it and to act quickly; take proper advice on anything you’re unsure about.
Not legal advice. Disclosed. is an information and template tool, not a law firm. The EU AI Act’s official text and the European Commission’s guidelines are the authoritative sources; the Commission’s guidance and Code of Practice were still being finalised close to the deadline. Nothing here creates a solicitor–client relationship, and it can’t account for your specific circumstances. For decisions with real consequences, consult a qualified adviser in your jurisdiction.