Skip to content
Atropox
Sign in

Legal documents · For everyone (AI Act, art. 50)

Artificial Intelligence Notice.

Where Atropox uses artificial intelligence, with what limits, with what safeguards for people, and what it never does.

Version 1.2 updated on 13 October 2026 7 min read

The full text

In short

  1. AI is used in the assistant (also by voice), in Intelligence Reports, in filling in contact cards, in finding photos and logos, in suggested tags and in reading documents. It only runs when you ask, and it is always flagged.
  2. The model is Anthropic’s Claude, which does not use your content to train its models. Voice goes through ElevenLabs, only if you choose to use it.
  3. AI proposes, you decide: it can make mistakes or mix up two people with the same name. Every data point has its source, to be checked.
  4. It never does social scoring, biometric recognition or inferences on sensitive data.
  5. It does not search for or profile minors, members of law enforcement, armed forces, the judiciary or intelligence services, or people who have objected. For people it reports no criminal data and no assessment of their solvency.
  6. Every report, PDF included, is marked as AI-generated. It is a limited-risk system under the AI Act; high-risk uses are forbidden by the Terms.

The summary helps you read. The full text below is what counts.

Questions about this document

Do you train AI on my data?

No. Anthropic, the model provider, does not train on anything we send it; NOK Group does not use your contacts’ data to train models and lists no training purpose among those declared in the Privacy Notice.

Where it is written Terms, arts. 5.7 and 7.3
What if the AI gets it wrong?

It can happen: it can mix up two people with the same name or use old sources. That is why every data point has its source, nothing is saved without your confirmation and AI output must always be checked.

Where it is written AI Notice, art. 5
Can I use a report to decide on a hire or a loan?

No, never, not even as a secondary factor. The Terms forbid it: these are high-risk uses under the AI Act.

Where it is written Terms, art. 5.4
All questions, your rights and your commitments →

The full text · Version 1.2

Issued pursuant to Article 50 of Regulation (EU) 2024/1689 (AI Act) and consistent with Law no. 132 of 23 September 2025

Version 1.2 — 13 October 2026

This is a courtesy English translation. In case of discrepancy, the Italian version prevails.

The Atropox service, offered by NoK Group S.r.l., integrates features based on artificial intelligence systems. This Notice describes such features, their purpose, their limits and the related safeguards, in fulfilment of the transparency obligations of art. 50 of Regulation (EU) 2024/1689 (the "AI Act"), as amended by Regulation (EU) 2026/1744, applicable from 2 August 2026 (for the marking of content of systems already on the market, from 2 December 2026), and consistent with Law no. 132 of 23 September 2025, in force from 10 October 2025, which for Italy designates AgID as the notifying authority and the National Cybersecurity Agency (ACN) as the market surveillance authority, without prejudice to the powers of the Data Protection Authority.

2. The AI features of the Service

2.1. Conversational assistant. The Service provides an assistant that understands the user's requests, responds and may propose or perform actions within the user's Network. The user is always informed of interacting with an AI system and not with a natural person (art. 50(1) AI Act).

2.2. Intelligence Reports (OSINT). At the user's specific request, the Service generates a documentary analysis of a subject indicated by the user (natural person or company), drawn exclusively from publicly accessible information found through web search. The report may contain a summary, synthetic indicators, a risk assessment, verifiable facts and the list of cited sources.

2.3. The features rely on the general-purpose AI model "Claude", provided by Anthropic, PBC, a company based in the United States, integrated into the Service. NoK Group integrates that model into the Atropox AI systems, of which it is the provider under the AI Act (downstream provider with respect to Anthropic's model), and uses them as a deployer for its own internal activity. The obligations relating to the general-purpose model rest with Anthropic, PBC. The necessary data is transmitted to Anthropic solely to process the requested response and is not used to train Anthropic's models, under the relevant commercial terms; transfer to the United States takes place on the basis of the Standard Contractual Clauses (Decision (EU) 2021/914). The marks "Claude" and "Anthropic" belong to Anthropic, PBC, an entity not affiliated with NoK Group, and are referred to for descriptive purposes only.

2.4. Activities the Service does not carry out. The Service does not perform social scoring, mass surveillance, biometric recognition or categorisation, inference of sensitive data, or automated profiling producing legal or similarly significant effects on persons. The Service does not generate reports, fill in cards or search for photos about minors: in such cases it returns a neutral outcome, with no personal data. Persons who hold or have held public office — members of law enforcement, the armed forces, the judiciary or intelligence services — are analysable: excluding them because of the office they hold would be discriminatory in effect and would make the politically-exposed-person checks required by anti-money-laundering rules impossible. For such subjects the analysis is confined to the public and professional dimension and in all cases excludes private address, household composition, routine and movements, vehicles, places frequented, operational duties and protection arrangements. It does not run searches about persons who have objected to processing (art. 21 GDPR). For natural persons, reports contain no data on criminal convictions, offences or criminal proceedings, and no assessment of creditworthiness or personal solvency. Assessments and scores are not based on protected attributes (gender, ethnic origin, religion, orientation, age) or their proxies. These limits are permanent and are maintained also in the future development of the Service.

2.5. Sources. Reports draw on publicly accessible sources and authority registers (sanctions, politically exposed persons, anti-corruption, business registers, official gazettes, insolvency proceedings, supervisory rolls), used to qualify an entity or for point compliance checks on a business counterparty, not to build the personal profile of an individual. For natural persons, compliance checks are limited to official restrictive measures (sanctions lists) and politically-exposed-person status.

2.6. Other artificial intelligence features. At the user's request the Service also uses the model to: a) fill in a contact card with public professional information found on the web, proposed with its source and applied only if the user accepts it; b) find a contact's photo or logo on public pages (never from social networks), proposed with its source and saved only if the user chooses it; c) suggest tags for contacts, which the user confirms or discards; d) read documents uploaded by the user to extract their text and propose actions, without keeping the file; e) summarise information pasted by the user. The safeguards of art. 2.4 apply to every search about a person.

2.7. Voice features. Using voice is optional. When the user speaks to the assistant, the audio is sent to ElevenLabs, Inc. (United States) for transcription only; when the user asks to hear a reply, the text is sent to the same provider to be read aloud. The transfer takes place on the basis of the Standard Contractual Clauses and the provider's EU-US Data Privacy Framework certification. If the provider's voice is not active or not available, dictation and the wake phrase («Buongiorno Atropox», which the user can switch on and off) use the speech recognition built into the user's browser: in that case the audio is processed by the browser's maker (for example Google for Chrome, Apple for Safari) under its own terms, and NoK Group receives only the transcribed text.

3. Risk classification under the AI Act

3.1. The features described are configured as limited-risk systems, subject to the transparency obligations of art. 50 AI Act, and not as high-risk systems under Annex III of the Regulation, as they are intended to support the organisation of the user's activity and provide summary information, not to decide autonomously about persons.

3.2. Use of the features for purposes that would make them high-risk systems — in particular for creditworthiness assessment, recruitment or personnel management, access to insurance or healthcare services, law enforcement, or other areas of Annex III of the AI Act — is prohibited by the Terms and falls under the user's sole responsibility, who in that case assumes the corresponding regulatory obligations.

4. Human decision support — no automated decision

4.1. Content generated by the AI features constitutes informational support for the user's decision and not an automated decision producing legal or similarly significant effects under art. 22 GDPR. The assessment and resulting action remain with the person.

4.2. The user shall exercise human oversight over the generated content, verify it at the source and not take significant decisions based solely on it.

5. Limits and reliability

5.1. Generated content may be inaccurate, incomplete, out of date or refer to a namesake. In particular, Intelligence Reports rely on public sources that may be wrong or dated. The absence of findings is not proof of the absence of information.

5.2. NoK Group does not warrant the accuracy, completeness or fitness for a particular purpose of the generated content, which does not constitute a certified check or professional due diligence.

6. Protection of personal data

6.1. The processing of personal data connected to the AI features is described in the Privacy Notice and, for third-party data contained in reports, in the Data Processing Agreement (DPA) under art. 28 GDPR. Intelligence Reports exclude by default the special categories of data (art. 9 GDPR) and, for natural persons, data on criminal convictions and offences (art. 10 GDPR). Reports on natural persons are deleted automatically 12 months after generation.

6.2. For the analysed subjects' data, not collected from the data subject, the information obligation under art. 14 GDPR and the identification of the legal basis (as a rule, legitimate interest, subject to a documented assessment) rest with the user as controller.

6.3. The AI model used is provided by a third party that processes data solely to deliver the service and not to train its own models, under the relevant commercial terms. Transfer of data to the United States takes place on the basis of the Standard Contractual Clauses (Decision (EU) 2021/914).

7. Literacy and responsible use

7.1. NoK Group takes measures to support the AI literacy of its staff and of the collaborators operating the systems (art. 4 AI Act) and provides users with the information needed for an informed use of the AI features and recommends using them in compliance with the principles of fairness, non-discrimination and protection of people's rights.

8. Contacts and updates

8.1. For information on the AI features, write to info@nok.business. This Notice may be updated; the current version is published in the Service with the date indicated.

Back to top ↑