Transparency in the use of artificial intelligence

1. What AI is used for on the platform

Artificial intelligence is used only in support functions for the documentary work of sessions:

• Transcription of session recordings.
• Speaker identification by voice, optional and only with each elected member's consent (see point 7).
• Proposed minutes, prepared from the transcript, the documents and the recorded attendance and votes.
• Reading scanned documents and handwritten paper notes.
• Short summaries of agenda items for the public portals.
• Meaning-based search across the entity's documents.

2. What AI does not do

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

AI output is a working tool: proposals and drafts that may contain errors, omissions or the wrong identification of a speaker. On their own they are not approved minutes, a decision, a legal opinion or an official record.

3. Human oversight

• Each proposed set of minutes is reviewed by authorised people, who can change, complete or reject it, checking it against the recording and the documents.
• Each generation is kept as a numbered version, separating what came from the machine from what people wrote.
• Every run records the stage, the engine used and the time.
• Minutes only become a document of the body once an identified person approves them, and that approval is recorded with name and time. Official status comes from the legal approval and signature procedure.

4. Roles under the AI Act

Regulation (EU) 2024/1689 distinguishes those who provide an AI system from those who use it. Olhar Panorâmico - Unipessoal, Lda. develops the system and makes it available under the Plenários brand: it is its provider. The entity that uses it, under its own authority, is the deployer. The general-purpose AI models the system relies on carry their own obligations, borne by the companies that provide them.

The Regulation entered into force on 1 August 2024 and applies in stages: prohibited practices and the AI literacy duty since 2 February 2025, the rules for general-purpose models since 2 August 2025 and the Article 50 transparency obligations since 2 August 2026.

5. Transparency towards readers

Article 50 requires certain AI-generated content to be disclosed. For text published to inform the public on matters of public interest, such as minutes, paragraph 4 waives that disclosure when the text has undergone human review or editorial control and a person holds editorial responsibility for the publication. That is what happens with minutes: editorial responsibility lies with the entity, and who approved them is recorded.

Even so, each portal publishes its own transparency page on the use of AI, and the platform keeps the records that document the review. Inside the platform, generated text is marked as such, in versions of its own, and the platform can preserve or add machine-readable provenance marks where required or technically appropriate.

6. AI literacy

Article 4 requires providers and deployers of AI systems to ensure their staff know them well enough to use them properly. The onboarding training for entity teams covers how AI works on the platform, its limits and how to review proposed minutes against the recording.

7. Speaker identification by voice

It is an optional feature, switched off by default. Only the entity can enable it, and it only works for elected members who have given explicit consent for their voiceprint to be stored. A voiceprint is biometric data (GDPR, Art. 9), and consent can be withdrawn at any time with the entity. Without that consent, an elected member's voice is not compared with any stored voiceprint.

8. AI providers and data

Transcription uses Gladia (France) and, as an alternative, Speechmatics (United Kingdom); speaker identification uses pyannoteAI (France); drafting and document analysis use models from Anthropic and OpenAI (United States). Data is sent only to produce the requested result, with the GDPR safeguards for transfers. The full register, with each provider's role, data involved and transfer mechanism, is in the Sub-processor register.

Entity data is not used by Olhar Panorâmico - Unipessoal, Lda. to train its own models, and AI providers are engaged on terms that prevent its use for general training, to the extent each service allows and guarantees.

9. Reporting an error

If you find an error in minutes or a summary published on a portal, report it to the entity that published it, through that portal's contact details, stating the session and the document. For questions about the platform, write to [email protected].

10. And on this site?

The plenarios.pt site does not use artificial intelligence to process the data of visitors or of people who fill in the form, and it has no automated chat assistants. The images of the platform are real screenshots.