Safety and privacy
C1 Ai is a spoken AI tutor used mostly by children. This page sets out, in one place, how we protect them. The binding texts are our privacy policy and terms.
1. Who can use it
- Only adults (18 or over) can create an account. Adults may study for themselves.
- Outside a class run by a school, a child studies only in a profile inside a parent's or guardian's account. At sign-up the adult confirms they are 18 or over and accepts the terms. Consent for a child's data is a separate step: when adding each child, the adult confirms they are the child's parent or guardian and consents to that child's data being processed for the lessons. The date and version of each consent are recorded.
- Before a child can be added or a lesson started, the adult confirms their email address through a link we send to it (a reasonable check under GDPR Art. 8(2) that the consent comes from the parent). Sign-in with Apple counts as confirmed, since Apple has verified the address.
- There is no way for a child to create an account or to start a lesson without one: our server refuses it.
- In a class run by a school, the school is responsible for its pupils' data and for informing parents, under its agreement with C1 Ai. Pupils join by the class's link under a pseudonym, with no account of a parent's and no password. See data protection and child safety for schools.
- Parents set limits per child: a daily time limit, a bedtime window, microphone on or off, slower speech and captions.
- For children who find reading or focusing hard, parents can turn on learning support: wider spacing, read-aloud, short one-step explanations and break reminders. It is based on published research, asks for no diagnosis and makes no medical claim.
- Outside a school class, a teacher sees a child's results only if the parent shares them, with a separate consent, by entering the teacher's class code in the child's profile. The teacher then sees, on the class's chosen topics only: the child's first name, time spent, before-and-after check scores, how often answers were right and which common misconceptions appeared. Never the lesson conversations. The parent can stop sharing at any time.
2. What the AI tutor does, and does not do
- It says it is an AI. The lesson screen always shows "you are talking with an AI, not a person" in the child's language (EU AI Act, Art. 50), and the tutor says so if asked.
- It keeps to the lesson. Its instructions forbid asking for personal details such as an address or school, steer away from topics unsuitable for children, and, if a child says they are being harmed, tell the child first to speak to a trusted adult (a parent or teacher), and, if they are in danger, to call the emergency number of the country they are in (for example 112 in the EU, 911 in the US), which the tutor names where we know it.
- It teaches, it does not judge. Marks, progress stars and quiz results are feedback for practice. They have no legal or similarly significant effect and are not used to decide anything about a child (GDPR Art. 22).
- No emotion recognition. The child's voice is turned into text so the tutor can answer. We do not infer feelings, mood or stress from the voice, face or words (prohibited in education by EU AI Act Art. 5(1)(f)). The tutor character's expressions are animations chosen for the lesson, not a reading of the child.
- Anything can be reported. Every lesson has a Report button; reports reach us with the last turns of the conversation so we can review them.
3. Where data goes
| Our servers | Hetzner, Helsinki (Finland, EU). Nightly encrypted backups in Germany, kept for at most 8 weeks. |
|---|---|
| Live voice | Google Gemini through Google Cloud Vertex AI, in the EU region europe-north1 (Finland), under the Google Cloud terms and data processing addendum. We do not use the consumer Gemini API, whose terms exclude services for under-18s. On this paid service Google does not use the data to train its models. |
| Text tasks | Anthropic Claude (following the lesson, marking answers, homework, quizzes), used in line with Anthropic's guidance for organisations serving minors: AI disclosure, child-safety instructions, and a way to report content. Anthropic's commercial API does not train on this data by default. Anthropic may process outside the EU under standard contractual clauses or the EU–US Data Privacy Framework. |
| Voice recordings | Not stored. The lesson transcript is kept while the account exists, and deleted after 12 months without that lesson being opened. |
| Uploaded books and photos | Private to the account that uploaded them (a teacher's own material chosen for a class can be studied by that class's shared pupils); read on our own server where possible. |
| Access to our systems | The admin dashboard needs an allow-listed account with two-step verification; the database is not reachable from the internet; the server accepts SSH keys only. |
| Advertising, profiling, selling data | None. |
| Usage statistics | Anonymous; nothing is stored on the device for them (no cookies). |
4. Deleting data
A child's profile, or the whole account with all its data, can be deleted at once from Settings → Privacy and data. A copy of all data can be downloaded there too. Deleted data leaves our backups within 8 weeks.
5. Risk assessment
We have carried out a data protection impact assessment (GDPR Art. 35) covering children's data, voice and the use of AI, together with a record of processing activities (Art. 30). It covers, among others: the AI providers' terms for services used by children, training on children's data, inappropriate AI output, data leaving the EU, loss of data, consent that does not come from a parent, and the EU AI Act (transparency, emotion recognition, and the rules for AI in education that apply from December 2027). We review it whenever the service changes; a summary is available on request.
6. Contact
Questions, concerns or a report about a child's data: support@c1ai.org. You can also complain to the Cyprus Commissioner for Personal Data Protection.