Teylio · Terms of usePrivacy policyCompliance and safety commitmentsPublic list of subprocessorsLegal notice · Françaisעברית
Compliance and safety commitments
v0.1 · 2026-09-06 · Working version drawn from Teylio's product and safety constitution; legal review in progress before access opens to families.
Five non-negotiable rules
R-01 The parent is the customer and account holder; the child never has their own login. R-02 No diagnosis, no label, no sensitive data for pedagogical purposes. R-03 The AI always says it is an AI and is never a friend. R-04 No engagement design (no streaks, no notifications to the child, no purchase on the child's side). R-05 Children's data is never sold, never used for advertising, never used to train a model.
Interaction safety
Every incoming message is neutralised then classified before reaching the mentor; every reply is checked before display. Prohibited content is refused with a return to the creation. Two-branch crisis protocol: risk to the child → session stopped, helpline shown, parent alerted immediately with the exact excerpt; danger possibly involving an adult of the household → session stopped, no alert to the household, human review within 24 hours under a written procedure. No interaction between children, no ranking.
What the child does themselves
The division of work is explicit: the child writes, decides, tests and signs their part; the AI helps, proposes, renders. One step of every mission is done without AI. Hints are graded and the child never receives a grade or a comparison.
Report to the parent
Weekly, in three sections (what they did, what they learned with the proof, the next step), based only on dated actions, without inference, without comparison to other children, without commercial recommendation. Every report passes a strict lexical check before sending.
Publication
No creation is visible outside the account without a link created by the parent, after a full scan (personal data, vocabulary, content, identifiers), on a separate, non-indexed origin, revocable at any time. Every newly published version is rescanned.
Checks before each release
Red-team in every active language, automated regression suite of the mission engine, vocabulary check on a sample of reports, update of this page, changelog. Sensitive features (Hebrew, publication, images, voice) stay closed behind switches until their acceptance is complete.
Feature switches (live)
- ATELIER_ETAPE_2 : closed
- BETA_INTAKE_ALLOWED : open
- BUILDER_MODE : closed
- CREATOR_UNIVERSE : open
- EXPLORER_MODE : closed
- HEBREW_NATIVE_REVIEW : open
- IMAGES_ENABLED : closed
- PILOT_EXPERIENCE_READY : open
- PUBLISHING_ENABLED : closed
- VOICE_ENABLED : closed