In short
Article 50 of the EU AI Act sets transparency rules for AI systems that talk to people or generate content. For AI-written text, the main duty falls on the provider of the AI system: its output must be marked in a machine-readable way as artificially generated.
The deployer — the business using the tool — must disclose AI-generated text when it is published to inform the public on matters of public interest. These rules apply from 2 August 2026.
For meeting notes, that usually means the tool's makers carry the marking obligation, and an internal summary is not the public-interest publication the deployer rule targets. Labelling AI-written notes where people read them is good practice anyway, for reasons that have nothing to do with the Act. This guide sets out what Article 50 says, who it applies to, and what a sensible label looks like.
What Article 50 says
The text is in Regulation (EU) 2024/1689, Article 50. Paraphrasing each paragraph that matters here:
| Paragraph | Who | What |
|---|---|---|
| 50(1) | Providers | AI systems that interact directly with people must be designed so people are told they are dealing with an AI, unless that is obvious |
| 50(2) | Providers | Systems that generate synthetic audio, image, video or text must mark the output in a machine-readable format, detectable as artificially generated |
| 50(4) | Deployers | Disclose deep fakes; and disclose AI-generated or manipulated text published to inform the public on matters of public interest |
| 50(5) | Both | The information must be given clearly and distinguishably, at the latest at the first interaction or exposure |
Two exceptions shape how this lands on meeting notes.
50(2) does not apply to assistive editing. The marking obligation does not apply to the extent the system performs an assistive function for standard editing, or does not substantially alter the input data or its meaning. Whether a meeting summary, which condenses a transcript into far fewer words, "substantially alters" it is exactly the kind of question the Commission's guidance and practice will settle. It is not settled here.
50(4)'s text rule has an editorial exception. Disclosure of published text is not required where the AI output has undergone human review or editorial control and a person holds editorial responsibility for the publication.
From when
The AI Act entered into force on 1 August 2024 and applies in stages. Article 50 applies from 2 August 2026.
In 2026 the EU adopted Regulation (EU) 2026/1744, the "Digital Omnibus on AI", which entered into force on 27 July 2026. It postponed the high-risk obligations. As summarised by several legal commentators (checked 25 September 2026), it left Article 50 on its original date, adding a grace period until 2 December 2026 for the machine-readable marking in Article 50(2) for systems placed on the market before 2 August 2026. Check the consolidated text on EUR-Lex for the current position.
The Act also asks the Commission to encourage codes of practice on detecting and labelling AI-generated content (Article 50(7)). The European Commission's AI Office publishes their status.
Who is who in a meeting notetaker
The Act defines roles by what you do, not by what you call yourself.
- The model's maker — the company that trained the language model — is a provider of a general-purpose AI model, with obligations of its own.
- The notetaker's maker builds an AI system on top of that model and puts it on the market under its name. It is likely to be a provider of that system for Article 50 purposes.
- You, using the notetaker at work, are likely a deployer. A person using it in a purely personal, non-professional activity is outside the definition of deployer.
If your meeting notes stay inside your team, the deployer's text obligation — about text published to inform the public on matters of public interest — is unlikely to be what you trip over. If you publish AI-written text externally, such as a public summary of a council meeting or a press release drafted from a transcript, look at Article 50(4) and its editorial exception more carefully.
Why label AI notes anyway
Set the law aside for a moment. There are three plain reasons to mark what the AI wrote.
Summaries go wrong in specific ways. A model can turn "we should look at that" into an action item with an owner, drop the one person who disagreed, or merge two numbers. Can you trust an AI meeting summary? goes through the patterns. A label tells the reader this is a draft to check, not a record someone vouched for.
Notes travel. A summary pasted into an email or a ticket loses its context. If the label goes with it, the next reader still knows where it came from.
Your edits should be distinguishable. Once someone corrects a summary, it is partly theirs. Marking which parts a person edited tells a later reader which parts someone has checked.
A good label is short, next to the text it describes, and survives copying: "AI-generated" beside each section, and "edited by you" once someone has changed it.
Related obligations to keep separate
The AI Act does not replace data protection law. Sending a transcript to an AI service is processing personal data, and the GDPR questions — lawful basis, notice, processors, retention — still apply. Recording meetings under GDPR covers them. Whether you may record the meeting at all is a third question again; see is it legal to record a meeting?
How Notey fits into this
Notey labels everything its AI writes, wherever it appears. This is a design decision, not a claim about any particular obligation under the Act.
- Each AI section is labelled "AI-generated" — the summary, action items, minutes, tough questions, where you differed, the follow-up email, and answers from Ask Notey.
- Your edits are marked. Change an AI section and it is shown as edited by you.
- The labels travel with exports. Markdown export to the clipboard or a file keeps the AI and edited labels.
- Only transcript text is sent to write the notes, when you ask or when a meeting ends with write-ups on, never audio. It goes to Notey's service and on to OpenAI as a processor, not used for training and not retained by Notey. The privacy policy has the detail.
- If the transcript does not cover a question, Ask Notey says so rather than guessing, and a summary with no commitments says so rather than inventing action items.
Frequently asked questions
When does Article 50 of the EU AI Act apply?
From 2 August 2026. The 2026 amending regulation known as the Digital Omnibus on AI moved the high-risk deadlines but, as summarised by several commentators, left Article 50 on that date, with extra time until 2 December 2026 for machine-readable marking by systems already on the market.
Do I have to label AI-generated meeting notes under the EU AI Act?
Article 50 puts the main text obligation on providers of the AI system, who must mark output as machine-detectable. Deployers must disclose AI-generated text published to inform the public on matters of public interest, which internal meeting notes usually are not. Whether a specific use is caught is a question for a lawyer.
Who is a provider and who is a deployer?
Broadly, the provider develops an AI system or places it on the market under its name; the deployer uses it in a professional activity. A company using a notetaker is usually a deployer; the notetaker's maker and the model's maker may each be providers.
Why label AI notes if the law does not require it?
Because the reader needs to know how much to trust them. A summary can state a commitment nobody made, and a label tells the reader to check it against the transcript before acting.