AI Transparency Notice

Provider / controller: ALGOPOETICA SPÓŁKA Z OGRANICZONĄ ODPOWIEDZIALNOŚCIĄ, Aleja Jana Pawła II 43A / 37B, 01-001 Warsaw, Poland · KRS: 0001036276 · NIP (Tax ID): 5273061434 · Contact: privacy@algopoetica.ai

1. Purpose of this document

This document fulfils the transparency obligation under Article 50 of Regulation (EU) 2024/1689 (the “AI Act”) — it explains where and how Algopulse uses artificial intelligence systems, their limitations, and the Provider’s role under the AI Act. It applies to both Users and Meeting Participants whose speech is processed by these systems. For information on the processing of personal data in connection with these AI systems, see the Privacy Policy — this document focuses on the AI systems themselves, not on personal data as such.

2. Where and how we use artificial intelligence

Function	What the AI does	Input	Output
Transcription	Converts speech to text	Meeting audio recording	Text transcript
Summary generation	Extracts the course of the Meeting, decisions, and action items from the Transcript	Transcript	Natural-language summary
Speaker attribution	Matches voice activity to audio tracks to label who is speaking	Meeting audio recording	Transcript with labeled speakers
AI Assistant answers	Answers the User’s questions about the User’s Customer Data, including Recordings shared with the User by other Users (under the User’s own Plan)	The User’s question and relevant parts of the Transcript	A natural-language answer; conversations are not visible to the people who shared the Recording

What the AI in the Services does not do: it does not converse with Meeting Participants (the Notetaker is not a chatbot — it joins a Meeting solely to record and transcribe it, not to ask or answer questions; the only conversation with an AI system is the User’s own, in the AI Assistant feature), it does not generate images or video, and it does not make decisions on the User’s behalf. Account registration, billing, and other features of the Services do not use AI systems.

3. Which models we use and where they operate

Transcription: Amazon Transcribe. Summary and AI Assistant answers: language models provided through Amazon Bedrock and run in AWS regions located in the European Union. Processing region: AWS cloud — regions located in the European Union. Amazon Web Services (infrastructure and model access via Amazon Bedrock) is a sub-processor under the DPA — the full list and transfer basis are set out in DPA Schedule C. Model developers do not receive data: the models are run by AWS. The specific language models may change; they always run in AWS regions located in the EU. A change of the infrastructure provider running the models, or moving processing outside the EU, will be reflected in an update to this document and notified to customers under Section 11.

4. Labeling AI-generated content

The Provider labels Summaries within the Services interface as machine-generated content. We do this as a transparency best practice, consistent with the purpose of Article 50 of the AI Act — even though the narrowest literal reading of that provision (which addresses content published to the public on matters of public interest) may not cover a Customer’s private, business Meeting summaries. The Notetaker is also visible in the Meeting participant list under a name identifying it as an Algopulse component (Algopulse | [inviting User’s email address]), which itself makes it obvious to a reasonably well-informed participant that they are interacting with an AI system before any Summary is even produced.

5. Limitations — what can go wrong

Transcripts and Summaries are generated automatically and may contain errors: transcription errors from poor audio quality, strong accents, industry jargon, or overlapping speech; misattributed statements, particularly with overlapping voices; omissions of matters discussed indistinctly or over background conversation; and hallucinations — a language model may generate a decision or action item in a Summary that was not actually said during the Meeting. This is a known limitation of language models, not a defect specific to any one Meeting — treat every Summary as a starting point for verification, not a finished record. Language limitations: transcription supports Polish and English — the Meeting language is identified automatically between the two. Meetings held in other languages, or switching frequently between languages, may be transcribed with significant errors, which also lowers the quality of the Summary. A Summary is not minutes, an official record, or evidence. Verify its content against the Recording or Transcript before relying on it for any purpose with legal or financial consequences.

6. Your data and model training

Recordings, Transcripts, and Summaries are not used to train, fine-tune, or improve any AI model — ours or any of our sub-processors’. This includes the models of providers we rely on (Amazon Transcribe and language models available through Amazon Bedrock). Your data is processed solely to generate a Transcript and Summary for your specific Meeting — it is never included in any shared training dataset. This commitment has no exceptions.

7. Human oversight and no automated decision-making

The AI systems used in the Services do not make decisions producing legal effects, or similarly significant effects, concerning you or any Meeting Participant on a fully automated basis (GDPR Art. 22). Transcripts and Summaries are informational — you evaluate their content and decide how to use it. See Section 10 to report an inaccurate Summary.

8. Classification under the EU AI Act

Provider’s role: under Art. 3(3) of the AI Act, the Provider is a “provider” of an AI system comprising the transcription and summary-generation functions described in Section 2. The developers of the language models are providers of general-purpose AI (GPAI) models accessed by the Provider via Amazon Bedrock (Amazon Web Services), without modification of those models.

Risk classification: limited-risk system under Art. 50, not a high-risk system. Reasoning: the Services’ AI functions are not listed in Annex III. The closest area is Annex III point 4(b) (decisions affecting employment conditions, including monitoring or evaluating employee performance or behavior) — the Services do not evaluate anyone’s performance or behavior, and do not support employment decisions; they produce only a Transcript and Summary of Meeting content. A Customer’s use of that output to evaluate employees would be a use inconsistent with the Services’ intended purpose — see the Recording Consent Guide (Section 8), which explicitly discourages this. The Services do not perform profiling under Art. 3(52) (automated evaluation of professional performance, economic situation, behavior, or reliability), so the Art. 6(3) profiling override — which would make a system high-risk regardless of other circumstances — does not apply. Speaker attribution is based on matching voice activity to audio tracks, not on comparing voice samples to identify a person — confirmed for Discord; the mechanism for Google Meet and Microsoft Teams (a single mixed stream) has not yet been verified. In neither case does the Service categorize individuals based on biometric data to infer protected characteristics (Annex III point 1(b)), so this function does not classify the Services as high-risk under that area either. This assessment is documented pursuant to Art. 6(4) of the AI Act, which requires providers to document a non-high-risk conclusion; this paragraph constitutes that documentation.

Applicable obligations and how we meet them (Art. 50): informing users of AI interaction (Art. 50(1)) — via the Notetaker’s visible presence in the Meeting participant list and the chat notice on Google Meet and Microsoft Teams (see the Recording Consent Guide, Section 7); labeling machine-generated content (Art. 50(2)) — via the Summary label in the Services interface (Section 4).

Status of Regulation (EU) 2026/1744 (Digital Omnibus) relative to the Services: Regulation (EU) 2026/1744, in force since 27 July 2026, deferred the deadlines for high-risk system obligations (Annex III: to 2 December 2027; Annex I: to 2 August 2028) — this deferral does not affect the Services, which were never subject to those obligations (see above). The Art. 50 transparency obligations that do apply to us were not deferred and have applied since 2 August 2026 — that is, already today.

9. What our system does not do

It does not analyze the emotion or mood of Meeting Participants — confirmed. Emotion recognition in a professional context is a prohibited practice under Art. 5(1)(f) of the AI Act; the Services do not perform it. It does not evaluate employee performance or behavior — see the classification reasoning in Section 8. It does not identify individuals based on biometric characteristics to infer protected attributes (race, political opinions, trade union membership, religious beliefs, sexual orientation) — speaker attribution is based on matching voice activity, not identity verification (Section 8). It does not build personality profiles of Participants or Users. It does not make, or automatically support, employment decisions. It does not score or rank Meeting Participants. It does not manipulate or use subliminal techniques to influence Participants’ behavior.

10. User rights and reporting problems

If a Transcript or Summary contains an error, you may request its correction under the general rules set out in the Privacy Policy (right to rectification, GDPR Art. 16, which also applies to data resulting from a malfunctioning AI system) — send your request to contact@algopoetica.ai. The Services do not currently offer a way to disable AI functions while retaining recording — transcription and Summary generation are an integral part of the Services; if you wish to record a Meeting without these functions, do not use the Notetaker.

11. Changes to our AI systems

We will update this document whenever we change our transcription or language model provider, the scope of processing performed by AI systems, or add a new AI-based feature. We will notify registered Users of material changes at least 14 days in advance, consistent with how we notify changes to the Privacy Policy. A version history of this document is retained and available on request.

12. Contact

Questions about the AI systems used in the Services: contact@algopoetica.ai. Data protection matters: privacy@algopoetica.ai. Provider’s address: Aleja Jana Pawla II 43A / 37B, 01-001 Warsaw, Poland.