Safety governance

AI should start with boundaries, not after-the-fact cleanup

Kids Protected AI supports scoped policies, intent classification, risk rules, tutor mode, sensitive-data handling, policy notices, and audit logs so schools can scale AI use with control.

Policy scopes 3

School, grade, and class policies can inherit or customize rules.

Actions 6

Allow, audit only, warn, tutor, redact, or block.

Events Auditable

Log user, session, rule, category, action, and excerpts.

Student protection Real time

Handle direct-answer, jailbreak, sensitive information, and unsafe image prompts.

Policy inheritance

Set whole-school standards, then tune by grade or class.

Administrators can control chat, attachments, image analysis, image generation, video generation, daily messages, daily cost, image limits, video-second limits, and response guidance modes.

School policy

Define the baseline for all students and teachers.

Grade policy

Adjust access and limits by age group or teaching plan.

Class policy

Customize rules for pilots, special courses, or class needs.

Reset inheritance

Return a scope to its default policy without rebuilding data.

Policy item Status Meaning
Text chat Enabled Students can use AI within authorized boundaries.
Daily cost Limited Control spend by class or grade.
Direct answers Tutor mode Convert answer-seeking into guided learning.
Sensitive info Redact/block Reduce exposure of personal data.
Risk handling

Safety rules can guide learning, not just block it.

For jailbreak, privacy, sensitive information, direct-answer cheating, and unsafe image prompts, Kids Protected AI can apply an action based on intent and severity.

Intent classification and risk categories

The system can classify student input into direct-answer, sensitive information, jailbreak, unsafe image prompt, and other risk categories before applying rules.

  • Support text and image-prompt risk checks
  • Record risk categories in policy events
  • Help teams tune rules and modes over time

Tutor mode

When students ask for final answers, AI switches to prompts, questions, and guided reasoning.

  • Preserve learning support
  • Reduce answer outsourcing
  • Keep events visible to teachers

Redaction and blocking

For sensitive data or high-risk content, the system can redact, warn, or block before provider calls.

  • Protect student information
  • Reduce unnecessary model exposure
  • Support audit and traceability

Safety audit

Policy events include user, session, message, rule, category, action, matched excerpts, and whether the model was called.

  • Leadership can review safety trends
  • Teachers can identify class risks
  • IT can debug rule matches

Policy notices

When a rule is triggered but learning can continue, the chat app can show a policy notice so students understand how the risk was handled.

  • Notices can work with tutor mode
  • Blocks and notices remain traceable
  • Teachers and admins can review events
Data governance

Local deployment lets schools design safety policy and data boundaries together.

For school-local or school-controlled environments, Kids Protected AI can align logs, attachments, reports, and model request flow with the school's governance requirements.

Explore local deployment
Governance goal Deployment support
Data control Plan data and log locations around school requirements.
Model access Use Gateway to centralize model APIs and availability.
Audit trail Keep policy events, usage, and reports traceable in the platform.