The AI Act and minutes: what has changed since 2 August 2026

Regulation (EU) 2024/1689, known as the AI Act, entered into force on 1 August 2024 and applies in stages. Prohibited practices and the AI literacy duty have applied since 2 February 2025. The rules for general-purpose models, since 2 August 2025. And since 2 August 2026, the transparency obligations of Article 50.

For a municipal council, an assembly or a parish board using artificial intelligence to prepare minutes, the question is simple: what does this change? This article sums up the essentials, based on how we work at Plenários.pt. It does not replace your organisation's own legal analysis, which belongs to your lawyers and your data protection officer.

Who is who under the Regulation

The Regulation distinguishes whoever supplies an AI system from whoever uses it.

• The provider is whoever develops the system and makes it available under their name. In the case of Plenários.pt, that is us.
• The deployer is the organisation that uses the system under its authority: the council, the assembly, the parish board.
• The general-purpose models the system uses inside have obligations of their own, borne by whoever provides them.

Knowing who is who helps you understand who answers for what, and ask each party for what is theirs to provide.

Article 50 and minutes

Article 50 requires certain AI-generated content to be identified as such. For text published to inform the public on matters of public interest, such as minutes, paragraph 4 provides an exception: identification is not required when the text has undergone human review or editorial control and a person holds editorial responsibility for the publication.

That is exactly how minutes work. Artificial intelligence proposes the text. An identified person reviews, corrects and approves it, and that approval is recorded with name and time. The body approves the minutes at the following session. Editorial responsibility lies with the organisation and is documented.

What matters is being able to show, at any moment, that there was human review and who took responsibility.

AI literacy: teams count too

Article 4 asks providers and deployers of AI systems to make sure their teams know them well enough to use them properly. In practice, whoever reviews a proposed set of minutes needs to know what the machine does well, where it tends to go wrong and how to check the text against the recording.

That is why the onboarding training for teams covers how AI works on the platform, its limits and how to review a proposed set of minutes.

What AI does not do

Knowing what AI does matters as much as knowing what is left out. On Plenários.pt, artificial intelligence:

• does not approve minutes or decide their official content;
• does not take decisions or produce administrative acts;
• does not vote or change votes or attendance, which are recorded by the chair;
• does not assess or rank people, recognise emotions or carry out biometric categorisation.

Speaker identification by voice is optional and switched off by default. Only the organisation can turn it on, and it only works for elected members who have given explicit consent, because a voiceprint is biometric data.

Six questions to ask whoever supplies your AI

Whether with Plenários.pt or another supplier, these questions help you see whether the system stands up to scrutiny:

1. Who approves the text, and is it recorded? The right answer is an identified person, with name and time.
2. Can we tell what came from the machine and what people wrote? Every generation and every revision should be kept as numbered versions.
3. Is every run recorded? Stage, engine used, time. Without that, there is no way to reconstruct what happened.
4. Is the content of our sessions used to train models? It should not be. On Plenários.pt, it is not.
5. Which AI providers are behind it, and where do they process data? They should be listed, with each one's role and location, and with the GDPR safeguards for transfers.
6. What happens when the AI does not know? A serious system flags the doubt instead of filling the gap with something plausible.

What stays with the organisation

The final legal qualification of the system belongs to the organisation and its data protection officer. The provider's role is to supply the record that supports that analysis: versions, approvals, runs, providers. On Plenários.pt, the principle predates 2 August 2026: AI proposes, a person approves, and everything is recorded.

The details are on our AI transparency and data protection pages. If your legal team has specific questions, get in touch.