AI Act Article 50 check: chatbots and generated content
Article 50 of the AI Act has applied since 2 August 2026, and it is the rule that affects the most companies: anyone with a chatbot on their website, a voice assistant on the phone, or who publishes text or images generated with artificial intelligence. You do not need a full audit to find out whether you comply. You need to look at those systems one by one, with the text of the regulation in hand, and write down what has to be fixed.
What Article 50 requires
These are transparency obligations: anyone who meets an artificial intelligence system must be able to know it. In practice there are four rules, and each one is addressed to someone specific.
- Chatbots and voice assistants (para. 1). People talking to one must be told that a machine is on the other side, unless it is obvious. The obligation falls on the provider of the system, but visitors see it on your website: if you had the system built for you and use it under your own name, the provider may be you.
- Generated content (para. 2). Providers of systems that generate text, images, audio or video must mark the output in a machine-readable format, so that it can be recognised as produced by AI.
- Emotion recognition and biometric categorisation (para. 3). Anyone using these systems must inform the people exposed to them.
- Deepfakes and text on matters of public interest (para. 4). Anyone publishing images, audio or video that imitate real people or events must disclose that they are artificial. The same applies to generated text published to inform the public on matters of general interest, unless a person has reviewed it and takes editorial responsibility for it.
The information must be given clearly, at the latest at the first interaction or exposure, and must be accessible to people with disabilities too (para. 5).
The two dates
Transparency towards people has applied since 2 August 2026, with no extension. For machine-readable marking, the amending regulation adopted in 2026 (the so-called Omnibus, Regulation (EU) 2026/1744) gave a few extra months only to generative systems already on the market before 2 August 2026: for them the deadline is 2 December 2026. A system placed on the market after 2 August must mark its content from the start.
Penalties for breaching these obligations reach 15 million euros or 3% of worldwide annual turnover (Art. 99). For SMEs, the lower of the two amounts applies.
What we check
- The list of systems that talk to people or produce content: the website chatbot, assistants on WhatsApp or on the phone, text and image generators used in marketing, tools that produce video or voices.
- Who is the provider and who the user of each one, because the obligations depend on the role, and what the contract with whoever sold it to you says.
- Where and when the system discloses itself: when it opens, during the conversation, in voice replies, in published text.
- Whether generated content comes out marked, and how: metadata, watermarks, visible statements. And what happens when the content is downloaded and republished elsewhere.
- Accessibility of the notice: whether people using a screen reader or a keyboard get the same information as everyone else.
What you receive
- A written report for each system: which paragraphs of Article 50 apply, whether they are met, with the evidence collected (screenshots, configuration extracts, contract clauses).
- The list of fixes, in order of urgency, with who has to make them: you, the system provider or us.
- A text for the provider, when the fix is theirs: what to ask for and on what basis.
The report uses the same format we use on ourselves. The chatbot on this website was checked this way: it discloses that it is artificial intelligence in three separate places in the panel, in the header, in the welcome message and in a notice that stays visible throughout the conversation.
When this check is enough and when you need the audit
If artificial intelligence in your company stops at a chatbot, a voice assistant or content production, the Article 50 check covers what concerns you today. If you use systems to select staff, assess customers or take decisions that affect people's rights, you may fall into the high-risk category, with much heavier obligations from 2 December 2027: that calls for the full AI Act audit, which includes Article 50.
Scope and cost are fixed and agreed beforehand, in a call, once we have counted together the systems to check. The report is written with a lawyer in the team.
Do you have a chatbot or publish generated content?
In a free 30-minute check-up we count the systems to check together and tell you what Article 50 means for you, before any commitment.
Book the free check-up