A school district proposes student-chat tutoring AI. What special consideration should audit map before model approval?

A school district proposing student-chat tutoring AI must map sensitivity and age-appropriate constraints for models that interact with minors before approval. Minors introduce heightened privacy, content-safety, logging, parental notice, and educational-purpose limits that differ from adult customer-service bots. Audit looks for age-appropriate filters, restricted tool use, retention minimization, prohibited topics, human escalation for distress signals, and alignment with student-data rules applicable to the district. Interaction logs may contain behavioral or academic signals that require stricter access control than ordinary help-desk transcripts. Relying on standard adult marketing survey anonymization practices for student chat logs understates sensitivity and likely legal constraints on educational records. Expanded tutoring topic coverage through third-party web browsing plugin integration prioritizes breadth over safety and creates data-exfiltration paths outside district-approved systems. Prior vendor demo safety testing as classroom content validation evidence confuses a controlled demonstration with live safeguards and does not reduce residual harm if the model generates inappropriate guidance. Exam caveat: “educational use” does not automatically authorize training on student chats or open plugin access; purpose limitation still applies. Operational check: verify DPIA/privacy review, age-gating and content policies, vendor subprocessors, retention and access logs, and a pilot limited to approved courses with guardian communication where required. Confirm blocked categories are tested with adversarial prompts before wider student rollout.