Skip to content

Artificial intelligence in 3S services

The 3S services, provided by “Demerzel” S.R.L., a limited liability company registered in the Republic of Moldova (its details are in the Legal notice), use an artificial intelligence (AI) assistant that answers questions about the Client's documents. This page says what the assistant does, what its limits are and what happens to the documents. It supplements the notice in the chat without replacing it: the assistant tells you that it is an AI system from the start of every conversation (section 2). The notice in the chat and this page take account of the transparency requirements of Article 50 of Regulation (EU) 2024/1689 (the Artificial Intelligence Act), to the extent that they apply. "Client" and "User" have the meaning given in the Terms and conditions.

In brief (a summary with no contractual value; the text below governs)

  • The assistant is an AI system, not a person. It tells you so from the start of every conversation.
  • Answers are generated automatically and may be incomplete or wrong. Check the cited document before relying on them.
  • Indicating the exact page of the source is in beta and may be missing or wrong.
  • Your documents are processed only to provide the service, under the data processing agreement (DPA).
  • You can always speak to a person: contact@3s.md or +373 60 055 599.

1. What the assistant is and what it does

The assistant is an AI system. It receives a User's question, searches the Client's documents and formulates an answer in text. It shows the sources on which the answer relies. It does not modify the documents.

Its purpose is searching the Client's documents and answering questions about them. It is not intended to evaluate people, for example in selecting candidates or assessing creditworthiness, and it does not take decisions in your place.

2. How you know you are talking to an AI system

The assistant tells you clearly, from the start, that it is an AI system and not a person: in the browser chat, through a message when the conversation opens.

This page supplements the notice in the chat; it does not replace it (Regulation (EU) 2024/1689, Art. 50(1) and (5)).

3. What is generated by AI

The text of the answers is generated automatically, on the basis of the question and the Client's documents. The documents themselves are not generated by the assistant and are not modified by it. An answer may summarize or reword the documents without reproducing them; treat it as generated text.

4. The limits of the answers

  • An answer may be incomplete, wrong or imprecisely worded. 3S does not guarantee that an answer is correct or complete (Terms, section 7.1).
  • The risk is higher with poor scans, handwriting, stamps and tables, figures, dates and proper names.
  • Text extracted from scans by optical character recognition may contain errors, and an extraction error can carry over into the answer.
  • The assistant sees only the documents uploaded and indexed in the Account. It cannot find what is missing from the archive, and the absence of a document from an answer does not prove that the document does not exist.
  • The same question, asked twice, may receive differently worded answers.
  • The assistant does not give legal, accounting or tax advice.

5. Sources and the exact page

The assistant indicates the documents on which it bases its answer. Indicating the exact page in the document is in beta: we are testing it together with clients, and the indication may be missing or wrong. Open the document and check before you rely on the answer (Terms, section 7.2).

6. Language

The result depends on the language of the documents, the language of the question and the quality of the scan. Questions in English about documents in Romanian are in beta.

7. What the Terms do not allow

The Terms (section 7.3) do not allow the assistant to be used, without 3S's written consent, as a component of a system intended for a use listed in Annex III to Regulation (EU) 2024/1689, for example selecting and evaluating job candidates or assessing the creditworthiness of natural persons, nor as the sole basis for a decision with legal effects on a person (Regulation (EU) 2016/679, Art. 22). The acceptable use rules are also in Notice and action, section 7.2.

8. Your documents and the AI models

AI models receive only text: the passages of the documents retrieved for the question or extracted from a specific document, the titles, the folder names, the metadata and the conversation history, and, during indexing, the text of the documents, in chunks. They receive it only to provide the service, under the Data Processing Agreement (DPA). The models and the text recognition are run by a sub-processor listed on the Platform sub-processors page; the model providers do not receive the documents. The agreement also contains 3S's commitment on the training of AI models (Art. 4.5). How the personal data in the documents is processed is described in the DPA and in the Privacy policy.

9. How to reach a person

You can always contact us directly, without going through the assistant: at contact@3s.md or on WhatsApp at +373 60 055 599 (messages and calls). A person replies, in English or Romanian.

10. How to report a problem

If the assistant gave an answer that is wrong, misleading or inappropriate, write to contact@3s.md, with the question, the answer and the date and time of the conversation; we will look into it. If you think an answer contains unlawful information, use the procedure on the Notice and action page.

11. What we guarantee and what we do not

Answers may be incomplete or wrong; the Client checks the cited document and is responsible for the decisions it takes on the basis of them. That is what the Terms say, in section 7, and the limits of liability are in section 13. This page adds no guarantees and adds no obligations for the Client.

12. Language of this page

The page exists in Romanian and in English. The Romanian version is the authentic one, and the English version is a translation; if the texts differ, the Romanian one applies.

Acts cited

The texts were read on September 30, 2026.

Text fingerprint (SHA-256)defe059ce738191e…