AI-Safety
AI Safety Guidelines
Gyanis uses a proprietary calibration engine to personalize learning across Flow, Vibe, Aura, and OneByZero. These guidelines ensure students and schools use our AI safely and responsibly.
Effective Date: December 16, 2025
Last Updated: December 16, 2025
Version: 1.1
Headquarters: Sharjah, United Arab Emirates • Domain: gyanis.ai
1) Core safety commitments
Age‑appropriate design
- Minor means under 18. Guardian consent is required where applicable.
- Age, board and class inform content level and tone.
- Teachers can request grade-based locks via Gyanis support.
Human in the loop
- Teachers and parents can review activity and reset personalization.
- “Ask an Expert” escalates to a verified teacher when AI is not enough.
Data minimization
- We collect only what is needed to provide the service and improve learning.
- Voice/transcripts are retained only as long as needed, then deleted or aggregated.
Safety escalation
- If we detect credible harm risks, we may notify guardians, schools or authorities as permitted by law.
2) Proper use - Students
- Use AI to understand concepts, practice MCQs, build mindmaps and plan study. Do not use AI to cheat on assignments or exams.
- Be clear and specific in prompts. If the answer seems wrong, ask for a simpler explanation or examples, or check with a teacher.
- Keep personal information private in chats and voice. Do not share phone numbers, addresses or passwords.
3) Proper use - Teachers
- Use Chat, Drills and Gameplan to differentiate instruction.
- Export drill PDFs for assessments. Review AI suggestions before sharing with a class.
- Adjust difficulty using the calibration controls. Flag unsafe or biased outputs; this feeds back into our safety systems.
4) Proper use - Aura Coaches
- Aura is for educational well-being, not diagnosis or emergency care.
- Keep session notes factual and age‑appropriate. Do not store medical records.
- Follow school and local safeguarding requirements. Report risks promptly.
5) Voice and recordings
- Vibe and live sessions require consent for audio processing. In two‑party consent regions, ensure all parties consent.
- No downloading of audio. Students can review session history where provided.
6) Sensitive content
- We block or soften content that is violent, explicit, hateful or otherwise inappropriate for minors.
- For mature topics in curricula, teachers can enable guided mode with stricter controls.
7) Academic integrity
- AI can draft outlines or explanations but final work must be the student’s own.
- Schools may set stricter policies; those policies apply on campus and in classwork.
8) Limitations & transparency
- AI may be incorrect or incomplete. Always verify important information.
- Our calibration engine adapts to learner signals, which can take time to stabilize.
- Recommendations such as YouTube videos are deep‑linked with full creator attribution; we do not control third‑party content.
9) Privacy & security basics
- Personalization signals can be reset from settings or by a school admin.
- We apply appropriate technical and organizational measures; no system is perfectly secure.
- For details, see our Privacy Policy and Terms on gyanis.ai.
10) Reporting issues
- In‑product: use “Report” on a message or output.
- Email: hello@gyanis.ai
- Schools: Gyanis support can enable stricter filters, disable features by grade, or reset cohorts.
Appendix - configuration for classrooms
- Feature toggles: enable/disable Vibe, Ask an Expert, and OneByZero listings for specific grades (via Gyanis support on teacher request).
- Content filters: default safe mode for minors; guided mode for mature syllabus topics.
- Audit: exportable logs for Chat/Drills/Gameplan activity; session summaries for Aura and Vibe available on request.
- Consent: guardian consent flag required for minors on Aura; voice consent prompts for Vibe.
- Reset: one‑click reset of personalization signals per class or cohort (by teacher, or via Gyanis support).
These guidelines refer only to Gyanis Educational AI and our proprietary calibration engine. They do not describe or expose any third‑party AI vendors.