AI Transparency

What models do we use?

Looping relies on large language models accessed via API. Since July 2026, quizzes, dictations, dialogues, and analyses are generated by Mistral AI, a French provider, in a digital sovereignty approach; OpenAI (GPT-4o) serves as a fallback solution in case of unavailability. The oral preparation in the Revision section relies on Anthropic (Claude). None of these models are trained on your children's data, and only the educational level, theme, and relevant curriculum excerpt are transmitted to them — never the student's name or identifier.

A Socratic AI, not an AI that does it for you

Our system instructions enforce the Socratic method: the AI never directly gives the answer a student must find. It guides in stages — targeted question, hint, partial explanation, then complete reasoning only after real attempts. Each correction explains the rule and reasoning.

Neither flattery nor false friend

Looping's AI has a neutral and encouraging tone, without automatic compliments or systematic validation. It does not simulate any emotional bond and is not designed to maximize time spent: it is an educational tool, not a companion. If a child expresses distress, it directs them to a trusted adult.

Its limits, honestly

Like any generative AI, it can make mistakes. We limit this risk (verification of responses, guidelines against inventing facts and quotes) and we give the student a right to challenge: each contested grade is reviewed and corrected if they are right.

Parental supervision

The parent creates the account, sets the rules, and can review their child's activity and interactions with the AI. Topics proposed by a child are subject to automatic moderation that excludes non-educational or inappropriate subjects.

Your data

Minimized data, hosted in the European Union, never resold, never used for advertising purposes. No third-party audience measurement tools in the child and parent areas.