Chinuch App uses AI models to help teachers and students with tasks such as generating lesson materials, drafting parent emails, transcribing speech, grading open-ended responses, and building remediation packets.
14.1 What we send to AI providers
Outbound AI requests follow two paths. They are not treated the same.
The assistant (the in-app chat, and texting a school number where that school has turned the text assistant on) works with school records only for a school that has accepted the AI Assistant Addendum, its authorization as the school. Until then it answers how-to questions only. For such a school it sends the names and school records the signed-in person is already allowed to open. That includes a student's name, and a parent's name or email when the task is to find that person or draft a message about them. Attendance, class, grades, and behavior-log text can be included on the same terms. The assistant also sends an internal id so it acts on the right record. Those ids are not meaningful outside our system.
Before an assistant request leaves our servers, a field scrub drops Social Security and tax identifiers, payment-card and bank numbers, Stripe identifiers, passwords, API keys, session tokens, and structured date-of-birth, street-address, medical, medication, allergy, diagnosis, IEP, 504, mental-health, and private-note fields. It also masks Social Security-shaped and card-shaped numbers inside remaining text. It does not replace names, emails, or phone numbers with placeholders. The assistant cannot find a person or draft a message about them if the name never leaves our servers.
Reports, narratives, and document tools (progress reports, remediation diagnosis, hearing reports, and similar generators) register the known student name and replace it with a placeholder such as "Student A" or "[Student]" before the prompt is sent. That text is also stripped of phone numbers, email addresses, street addresses, dates of birth, 9-digit school IDs, and card-like numbers. Where the finished text needs the real name, we put it back on our server after the model responds.
14.2 What we do not send to AI providers
- Payment card numbers, bank-account numbers, or Stripe secrets. Payments stay with Stripe.
- Social Security numbers and similar government identifiers.
- Passwords, API keys, and session tokens.
- A child's photo.
- Structured date-of-birth, street-address, medical, medication, allergy, diagnosis, IEP, 504, mental-health, and private-note fields.
- On report and document tools, the child's name, email, phone number, street address, date of birth, and school ID, which those tools replace or strip before the prompt is sent.
A behavior or health log's free-text description can still be included when a staff member asks the assistant about a log they are already allowed to read. The assistant does not use that text for a disciplinary decision on its own. A person confirms any write.
14.3 Child-safety guardrails on every AI call
- Every AI prompt that can reach a child is wrapped with a non-negotiable safety preamble that forbids sexual, violent, self-harm, substance-use, weapons, political, and age-inappropriate content, in line with Anthropic's guidance for organizations serving minors and equivalent OpenAI and Google policies.
- AI-generated content shown to a student is visibly labeled as "AI-generated" and includes a way to report inappropriate output. Reports are reviewed by our safety team and the child's school.
- AI is not used for disciplinary decisions, identity verification, mental-health diagnosis, or anything that permanently affects a child's record without a human reviewing it first.
For additional information specific to children under 13, see our COPPA Notice.