Artificial intelligence terms

1Nature of these terms

These AI Terms form part of the Sinapsi Contract between Celebron S.r.l. and the Client and govern the use of artificial intelligence in the service. On the matters they cover they prevail over the Sinapsi Terms of Service; for personal data the Data Processing Agreement prevails. Capitalised terms have the meaning given in the Terms of Service.

They implement Regulation (EU) 2024/1689 on artificial intelligence (the “AI Act”), Italian Law no. 132 of 23 September 2025 and Regulation (EU) 2016/679 (GDPR). They remain valid even if the classification of Sinapsi changes (Article 14).

2Roles

Celebron is the provider of the Sinapsi artificial intelligence system and uses it to provide the service. The Client uses the Dashboard under its own authority and is the deployer of the outputs it receives (Article 3 of the AI Act). General-purpose AI models are components supplied by third-party providers listed in the Data Processing Agreement: Celebron selects them, controls them and is liable for them to the Client.

3Intended purpose

Sinapsi serves only to support the management of the Client’s own business: it reads and links documents, represents accounts, cash and deadlines, makes forecasts and proposes actions, which a Celebron person verifies and the Client decides on. It is a decision-support tool: it does not decide and does not replace the judgement of the business owner or of its professionals.

In the app Sinapsi’s artificial intelligence appears under the trade name «Calypso» and with a graphic character, which may change without changing the service or these Terms. Calypso always states that it is an artificial intelligence. The game features it accompanies, such as the tower, missions, streaks and badges, serve only to guide the use of the service: they are not an assessment of the business or of its creditworthiness and give no prizes or economic benefits. In quiet mode the Client reduces Calypso’s messages and switches the game off.

4No acts towards third parties

Sinapsi’s artificial intelligence systems do not act on the Client’s behalf. In particular, they do not:

• make payments or give instructions to banks;
• send communications, certified emails, applications or filings to third parties or authorities;
• sign documents or accept or reject offers;
• change the Client’s accounts or official records.

Every act towards the outside world is performed by the Client or by the people it authorises.

5Prohibited uses

The Client does not use Sinapsi or its outputs to:

• assess the creditworthiness of natural persons or assign them credit scores, including shareholders, guarantors, employees and customers who are natural persons;
• take or prepare decisions on hiring, tasks, promotion, evaluation or dismissal of workers;
• profile natural persons or infer their personal characteristics or special categories of data;
• present to banks, authorities or third parties data it knows to be untrue or incomplete;
• any practice prohibited by Article 5 of the AI Act or otherwise contrary to law.

These uses fall outside the intended purpose. If the Client engages in them it is liable for them and, where this turns Sinapsi into a high-risk system, it may be considered a provider, with the related obligations, under Article 25(1)(c) of the AI Act. Celebron may suspend the access concerned, with written notice.

6Human oversight

Every forecast, proposal, deadline or line that reaches the Client as verified has been checked by a Celebron person with the appropriate expertise, whose role is shown in the Dashboard. The reviewer can correct, discard or block an AI output. Deadlines of legal documents are checked twice, by Celebron and by the qualified professional. Anything not yet verified is labelled as such.

7Quality of data and outputs

• Celebron checks that documents are complete and consistent and that every figure in the Dashboard is linked to the document it comes from.
• Forecasts are shown with a range and a probability, never as a single figure.
• At least every quarter Celebron compares forecasts with actual results and, on request, reports accuracy indicators to the Client.
• When it finds an error it corrects it and informs the Client if the error may have affected one of its decisions.

8Security and protection against manipulation

Celebron protects data and outputs against attempts to alter them, including through artificial intelligence. In particular:

• documents and emails received are treated only as data, never as instructions to the AI systems; content that tries to give instructions or alter results is isolated and flagged;
• every figure is reconciled with the source documents, and anomalous changes are verified by a person before appearing in the Dashboard;
• models receive only the data needed for each task, pseudonymised where possible;
• only authorised staff access the systems, with multi-factor authentication and least privilege, and each Client’s data are kept separate;
• outputs and logs are protected against unauthorised changes, with integrity checks;
• at least once a year and at every significant change Celebron tests the robustness and security of the AI chain, including through simulated attacks.

The Client safeguards its access codes and sends documents only from authorised addresses, as the Terms of Service provide.

9Logs and traceability

For every AI output Celebron records the source documents, the date and time, the model and version used, who verified the result and the outcome of the check. During the Contract logs are kept for at least twelve months; on termination the deletion rules of the Terms of Service apply, unless the logs are needed for a dispute. The Client may request an extract concerning its own data.

10Explanations and review

For any output the Client may request a plain-language explanation of the sources, assumptions and main factors, and a review by a person other than the one who verified it. Celebron replies within ten working days, or within two when the request concerns an imminent deadline.

11Models and providers

Celebron uses general-purpose AI models from leading providers, under professional terms that exclude training on customer data, limit retention and safeguard transfers as provided in the Data Processing Agreement. Before adopting a new model or a significant new version, Celebron tests it against Sinapsi’s checks, including accuracy, security and deadline calculation, and does not put it into use if it performs worse. Celebron does not use Client data to train third-party models.

12AI incidents

An incident is any error, malfunction, manipulation or misuse of the AI systems that has affected or may affect the Client’s data or outputs, for example a wrong deadline, an altered figure or data exposed to someone who should not see them. In the event of an incident Celebron:

• notifies the Client within forty-eight hours of discovery, stating what happened, which outputs are involved and what it has done;
• corrects the outputs and tells the Client what to check;
• analyses the cause, records it and takes measures to prevent it from happening again;
• makes the notifications to authorities required by law and assists the Client with its own, including those on personal data breaches.

13Continuity and operation without AI

If a model or provider becomes unavailable, unsafe or no longer lawfully usable, for example because of outages, security risks, decisions of authorities or export restrictions, Celebron switches to an alternative provider from the list of sub-processors, with equivalent safeguards, or continues the essential functions, namely legal deadlines, alerts and Dashboard data, through the team’s work without artificial intelligence, and informs the Client.

If the service cannot be maintained at a substantially equivalent level for more than thirty days, either party may withdraw without cost, with a refund of the fees paid for the period not used.

14Classification and changes to the rules

Celebron has assessed Sinapsi under Article 6 and Annex III of the AI Act and has classified it as not high-risk, for the reasons explained on the page How we use artificial intelligence. The assessment is documented and repeated before any new feature and whenever the law, the European Commission’s guidelines or the decisions of authorities change.

As a precaution, Celebron already applies to Sinapsi the main safeguards required of high-risk systems: human oversight, data quality, logs, accuracy, robustness and security, and information to the Client.

If Sinapsi or one of its functions were to be considered high-risk, by law, by decision of a competent authority or by Celebron’s own assessment:

• Celebron informs the Client within fifteen days;
• it brings the system into line with the requirements for high-risk systems, including risk management, technical documentation, a quality management system, conformity assessment and registration, within the deadlines set by law and at its own expense;
• in the meantime it may suspend or limit the function concerned or provide it without artificial intelligence;
• it gives the Client the instructions and information needed for the deployer’s obligations, which the Client complies with: use in accordance with the instructions, oversight entrusted to competent people, retention of the logs under its control for at least six months and information to workers where required;
• if the changes substantially alter the service or its price, the Client may withdraw without cost.

15Liability

AI outputs are estimates and proposals: decisions remain with the Client and its professionals. Celebron is liable for errors of its AI systems as for the rest of the service, with the diligence required by Article 1176(2) of the Italian Civil Code and within the limits of Article 15 of the Terms of Service, which do not apply in case of wilful misconduct or gross negligence.

Celebron is not liable for errors caused by untrue, incomplete or late documents, by uses outside the intended purpose or by failure to instruct the qualified professional in time.

Celebron maintains professional liability and cyber-risk insurance appropriate to the service; the limit is stated in the individual contract.

16Banks and other third parties

When the Client gives banks or other third parties access to the Dashboard, the Dashboard states that it is a representation prepared for the Client with the help of artificial intelligence and verified by Celebron, and not an attestation, certification or rating. The Client informs third parties of this limit. Third parties acquire no rights against Celebron without a written engagement.

17People and skills

Celebron ensures that everyone working on Sinapsi has adequate training on artificial intelligence, its limits and its risks. It appoints an AI Officer who oversees compliance with these AI Terms, the tests, the incident register and the classification assessments, and gives the Client the officer’s name on request.

18Audits

Once a year the Client may request a summary of the checks carried out: tests, accuracy indicators and incidents, with the measures taken. Audits at Celebron follow the rules of the Data Processing Agreement.

19Clauses specifically approved

Under Articles 1341 and 1342 of the Italian Civil Code, the Client specifically approves, by signing the individual contract, the clauses of the following articles: 5 (prohibited uses, suspension of access and provider obligations); 13 (withdrawal for unavailability of artificial intelligence); 14 (suspension or limitation of functions and withdrawal following a change of classification); 15 (limitations of liability).

The Italian text prevails over translations.