The EU AI Act transparency rules under Article 50 start applying on 2 August 2026. If your business uses AI tools with customers, staff or the public, there are a small number of things you now have to tell people. This is the plain version, with a checklist you can finish in an afternoon.
The short version
Article 50 is about honesty, not paperwork. It asks you to be clear when a person is dealing with a machine rather than a human, and to be clear when content has been generated or altered by software. That is the whole idea.
The three things that actually apply to most businesses
1. Tell people when they are talking to a bot
If you run a chat widget, a phone answering service, or a support flow that answers customers automatically, the person must be able to tell it is not a human. You do not need legal wording. A line such as "You are chatting with our automated assistant. Ask for a person any time." is enough. The rule only bites where it would not already be obvious.
2. Disclose synthetic images, audio and video
If you publish an image, voice clip or video that was generated or altered by software and it looks or sounds real, say so. Typical examples are a generated product scene, a cloned voiceover, or a face swapped into a video. A visible caption or an on screen label works. Obvious illustrations, cartoons and clearly artistic work are not the target here.
3. Label generated text only when it informs the public on matters of public interest
This is the part most businesses get wrong. Ordinary marketing copy, product pages, proposals, internal documents and customer emails do not need a label, even if software helped write them. The rule targets published text that informs the public on matters of public interest, and even then there is an exemption where a named person took editorial responsibility and reviewed the content before publication.
What you do not need to do
- You do not need to stamp every blog post, newsletter or slide deck with a notice.
- You do not need to disclose that software helped draft an internal report.
- You do not need a new policy document to comply with this specific rule, though having one makes it easier to prove.
Who is responsible
Using a compliant vendor does not transfer the duty. If your business puts the chatbot on your site or publishes the video, the disclosure is your responsibility. Vendors can help you do it, but they cannot do it for you.
Your checklist for this week
- List every customer facing automation. Chat widgets, voice answering, automated replies in email or WhatsApp, booking assistants. Write them down.
- Add a one line notice to each. State that the reply is automated and offer a route to a person.
- Check your published media. Look at the last six months of images, audio and video. Caption anything generated or altered that could pass for real.
- Agree one rule for published text. Decide who signs off content before it goes out, and record that person by name.
- Keep the evidence. Screenshots of the notices, the list of automations, and the name of your sign off owner. That is what an auditor, insurer or customer will ask for.
Why bother beyond the law
Buyers and procurement teams are already asking suppliers how they govern their use of AI. Being able to answer in one page, with dates and named owners, wins contracts. The disclosure work here is small, and it is the cheapest part of building that answer.
Where EUQA helps
Inside EUQA, the AI Inventory captures every tool and where it touches customers, the AI Policy module records your sign off rule, and the Evidence Log timestamps the whole lot so you can hand over proof without digging through folders.
If you want a starting point, run the free diagnostic and you will see which of these five steps you already pass.