EURO.SUPPORT / LEGAL DOCUMENTS

Responsible AI Usage Policy

Rules for transparent and safe AI customer support in euro.support

1. Status and purpose of the document

This Responsible AI Usage Policy (the “AI Policy”) forms part of the agreement for the euro.support Service. It governs AI translations, response suggestions, summaries, categorization, searches of knowledge sources, the AI agent, and AI-supported commercial functions. Its purpose is to provide international support in which AI helps overcome language barriers while remaining subject to appropriate human oversight.

2. Roles of the parties

The Provider designs and operates the Service’s AI functions and integrates third-party models. The Client decides which functions to enable, what knowledge and instructions to provide, to whom to allow automation, and how to inform and support End Users. In a specific deployment, the Client is the party using AI in its own process and is familiar with its context and risks.

3. Transparency towards people

If AI communicates directly with an End User and this fact is not obvious from the circumstances, the Client shall provide clear and distinguishable information no later than at the first contact. The Widget and Service settings must enable this information to be provided; the Client must not remove it or phrase it misleadingly. The user must have an understandable option to request a human agent, if available.

If the Client publishes AI-generated or substantially AI-modified content to inform the public, it is responsible for labelling and human editorial review in accordance with applicable law. Ordinary translation, wording adjustments, or an agent-reviewed response shall be assessed according to the specific context.

4. Accuracy, translations, and limitations

AI is probabilistic and may produce factually incorrect, incomplete, outdated, culturally inappropriate, or ambiguous output. A translation may change the meaning, tone, currency, date, legal term, or product name. The Client shall not present AI output as guaranteed to be accurate merely because it was generated by the Service.

• Significant responses must be based on the Client’s current and authorized knowledge sources.
• Price, payment, claim, order, cancellation, medical, or legal information must be appropriately verified by a human or reliable system data before taking effect.
• The Client shall configure escalation for low confidence, an unknown question, a complaint, a threat, sensitive data, or a request for a human.
• The agent distinguishes the original message from the translation and may review the original text where necessary.

5. Human oversight and commercial actions

The level of oversight must correspond to the risk. An agent must review a response suggestion before it is sent if it could create an obligation, reject a claim, change an order, or significantly affect a person. A fully automated response is primarily suitable for clear, low-risk, and verifiable questions.

A function that writes to an external store must have appropriate authorization and a confirmation step. The Client is responsible for cancellation and refund rules, availability, and customer rights. AI must not bypass technical confirmation or perform any action other than the one expressly authorized by the Client.

6. Prohibited and inappropriate use

AI functions must not be used for:

• manipulative or deceptive techniques that materially impair a person’s ability to make a free decision and may cause significant harm;
• exploiting a person’s vulnerability due to age, disability, or social or economic circumstances;
• social scoring, unauthorized profiling, or discrimination;
• biometric categorization, emotion recognition, or remote biometric identification; the Service is not intended for these purposes;
• solely automated decisions with legal or similarly significant effects without a valid legal basis, the required information, and the possibility of human intervention;
• professional decisions in healthcare, law, credit, employment, insurance, or critical infrastructure without a separate agreement, assessment, and compliance;
• generating unlawful, harmful, fraudulent, or content that infringes third-party rights.

7. Inputs, outputs, and rights

The Client is responsible for the lawfulness and quality of inputs, instructions, and knowledge sources. To the extent permitted by law, AI output belongs to the Client under the Service Terms, but it may not be unique, and a similar output may be provided to another user. The Provider does not guarantee copyright protection or non-infringement of third-party rights; the Client shall assess the risk before public or commercial use.

8. Data, training, and AI providers

The Provider does not use Client Content to train its own general-purpose model and, without the Client’s express instruction, does not enable voluntary sharing of content for training an external model. In standard mode, the OpenAI API does not use inputs or outputs for training unless the API customer expressly opts to share them. The Provider may temporarily retain limited security logs in accordance with its terms.

For searches, the Service may create embeddings and store them in a separate Client index. This is a technical representation of content for carrying out the Client’s instruction, not training of a general-purpose model. The processing of personal data is governed by the DPA and the list of Subprocessors.

9. AI literacy and the Client’s internal rules

The Client shall ensure an appropriate level of AI literacy among agents, administrators, and persons who configure or oversee AI. Training shall take into account their experience, the degree of automation, languages, vulnerable groups, and possible consequences. At a minimum, it must cover model limitations, verification of facts and translations, data protection, escalation, safe instructions, and recognition of prohibited use.

10. Monitoring, incidents, and feedback

The Provider may, to a reasonable extent, monitor technical quality, error rates, usage, security signals, and aggregated results. The Client shall report serious or recurring incorrect output, bypassing of confirmation, discriminatory patterns, or an incident to info@euro.support and preserve the necessary context. The Provider may temporarily restrict the affected function, modify an instruction, model, or security rule, and inform affected Clients.

11. Changes to models and functions

The model, its version, price, speed, or behavior may change. The Provider shall carry out appropriate testing of a significant change and shall not intentionally reduce the baseline level of protection. Following a material change, the Client should review its own automations, knowledge, and oversight. Beta AI functions are provided for testing and should not be used for high-risk purposes.

12. Contact

Questions about an AI function, transparency, security, or data processing may be sent to info@euro.support. The Client’s legal and security obligations in a specific deployment are not thereby transferred to the Provider.