Effective / last updated: September 20, 2026. This policy describes the safeguards we have built and the ones we are actively finalizing. Where a control is still in progress, we say so explicitly rather than claim it is complete. For the full picture of what data we collect, who we share it with, and your rights over it, see our Privacy Policy.
1. K-12 Student Safety Core™
Unlike general-purpose generative AI tools that operate open-ended queries, Khypri AI processes all exchanges through a proprietary 7-layer Safety Core™ system. This is designed to prevent content hallucinations, curriculum escapes, and student exposure to non-academic topics, supporting a safe, age-appropriate learning sandbox for K-12 students worldwide. The description below reflects our current architecture, which continues to evolve; no automated safety system can guarantee the complete prevention of every possible unsafe output.
- Layer 1 — Input Filtering: Incoming prompts are screened before reaching the AI model, flagging attempts to request non-academic, unsafe, or manipulative content.
- Layer 2 — Curriculum Boundary Check: Requests are checked against the relevant curriculum scope (CBSE, ICSE, IB, Cambridge, and supported State Boards) so the platform stays within academic subject matter.
- Layer 3 — Content Moderation: Automated classifiers flag violent, sexual, hateful, or otherwise harmful content before it can be generated or returned.
- Layer 4 — Age-Appropriateness Filter: Output complexity and tone are adjusted based on the student's declared grade level.
- Layer 5 — Accuracy & Hallucination Check: Generated academic content is checked against curriculum-aligned reference material to reduce factual errors, though it cannot eliminate them entirely (see the assumption-of-risk provisions in our Terms of Service).
- Layer 6 — Human-in-the-Loop Review: A sample of outputs, along with anything flagged by earlier layers, is reviewed by human moderators to identify gaps and retrain the automated filters.
- Layer 7 — Final Output Scan: A last automated pass runs immediately before content is returned to the student, to catch anything earlier layers may have missed.
2. Olli AI Tutor & File Uploads
Olli, our open-ended AI tutor, and the ability to upload notes/worksheets are both subject to the same 7-layer Safety Core described above, plus the following additional safeguards specific to open-ended input and file content:
- Upload Safety Check: Any file a student uploads is checked for safety before the student can use it or generate anything from it. A file that fails this check is not made available to the student.
- PII Redaction: Our default design avoids sending personal data to any third-party AI model in the first place. As an additional safeguard in case personal information is inadvertently included in a chat prompt or an uploaded file, we apply PII redaction to strip it out before any content is passed to a third-party model for processing.
- Topic & Content Guardrails: The same curriculum-boundary, content-moderation, and output-scanning layers that apply to generated practice papers also apply to Olli's open-ended chat and to any content derived from an uploaded file.
- Logging: Chat prompts, uploads, and Olli's responses are logged internally, so that a specific interaction can be reviewed if a safety concern is raised.
- Parental File Vault Controls: Parents/guardians can grant or revoke their child's access to the file upload/library feature at any time.
- Zero Outbound Hyperlinks & Walled Garden: Olli and Lumi operate as closed-loop, self-contained study companions. The AI system is programmatically barred from generating, suggesting, or displaying external hyperlinks, search engine URLs, or web addresses. Students are never redirected to unverified external internet destinations.
- Zero External AI Referrals & Model Isolation: The AI tutor is strictly prohibited from referencing, recommending, or providing access links to external third-party AI systems (such as Google AI, Gemini, ChatGPT, or external conversational bots). All educational interactions remain completely self-contained within Khypri, preventing exposure to external commercial platforms, cookies, or unmonitored tracking.
2.1 Automated Safety Detection & Parent Alerts
Olli's conversations with a child are automatically checked for signs of serious risk to the child's safety or wellbeing — for example self-harm, abuse, violence, bullying or severe distress. Where an automated safety detection flags potential sexual offences or grooming, it is specifically categorized under Section 19 of the POCSO Act, 2012 and immediately routed to a dedicated, high-priority escalation path rather than generic high-risk moderation. When a serious concern is detected:
- Supportive response: Olli responds with calm, supportive guidance and encourages the child to talk to someone they trust; the conversation may be paused.
- Immediate parent alert: the parent/guardian is alerted straight away by WhatsApp and email, with the severity and the type of concern.
- Reminders and a short summary: if the alert is not acknowledged we send further reminders over the following hours. Roughly 15 minutes after the first alert we also send a short summary — one or two sentences, written with the help of an AI provider listed on our Sub-Processors page — so the parent can check in with their child.
- Acknowledgement and support: a parent can acknowledge an alert from the WhatsApp message or the platform, and our support team may contact the parent about it.
- Record: the concern, the alerts we sent and whether they were acknowledged are logged internally, so the process can be audited.
These alerts go only to the mobile number and email address on the parent account, under the required “Essential alerts” consent and to protect a child who may be at risk of serious harm. Detection is automated and imperfect: it may miss a concern or raise one that turns out to be nothing, and it is an additional safeguard, not a substitute for a parent's own supervision or for emergency services. If a child is in immediate danger, contact your local emergency number (in India, 112) or Childline (1098). See Section 4.3 of our Privacy Policy for how this data is handled.
3. Content Reporting & Escalation
If a parent, teacher, or student encounters content that seems unsafe, inaccurate, or inappropriate, it can be reported from within the platform or by emailing support@khypri.com with the content in question. We aim to acknowledge safety reports within 24 hours and to review and, where warranted, remove or correct the flagged content within 5 business days. Reports involving a credible safety risk to a child are prioritized for immediate review.
4. Mandatory Reporting of Offences Against Children (POCSO Act, India)
This section is separate from general content moderation in Section 3 above. It exists because Indian law imposes a specific, personal, non-optional duty here: under Section 19 of the Protection of Children from Sexual Offences (POCSO) Act, 2012, any person — including anyone who works for or with Khypri AI — who has knowledge, or a reasonable apprehension, that a sexual offence against a child has been, is being, or is about to be committed, must report it. Under Section 21 of the POCSO Act, failing to do so is itself a criminal offence, punishable independently of whatever underlying offence was not reported. This duty applies the moment there is a reasonable apprehension — it does not require proof, certainty, or an internal investigation first.
4.1 Who Must Report
Every Khypri AI employee, contractor, content moderator, teacher-partner, and support staff member with any access to student interactions, uploaded files, chat logs, or communications is individually bound by this duty. It cannot be delegated away by assuming "someone else will report it," and it is not satisfied merely by flagging content for routine moderation review under Section 3 — a suspected child sexual offence must go through the pathway below, not the general content-report queue.
4.2 Who to Report To
A report must be made to the Special Juvenile Police Unit (SJPU) or the local police, as POCSO Section 19 requires. In parallel — not instead of — the reporting individual must immediately notify Khypri AI's internal escalation contact:
- Internal escalation: our Data Protection Officer and designated Child Safety Lead, Debasmita Maharana (or Deputy DPO / Compliance Escalation Team), reachable at dpo@khypri.com and, for anything urgent, via Call at +91 88679 64542. This contact is available for staff to escalate immediately, any time, not only during business hours.
- External, statutory reporting: the nearest Special Juvenile Police Unit or police station, and/or Childline India, 1098 (a free, 24/7 national child helpline), which can also guide a reporting individual through the process.
A report to Khypri AI's internal contact does not substitute for the statutory report to the SJPU/police required by law — both must happen. Our internal contact exists so that Khypri AI can simultaneously take platform-level action (see 4.4) and support the reporting individual, not to gatekeep or filter what gets reported externally.
4.3 Preserving Evidence & DPDP Erasure Override
Once there is a reasonable apprehension of an offence, anyone with access to relevant material must preserve it, unaltered, for law enforcement:
- Do not delete, edit, or "clean up" the chat logs, uploaded files, timestamps, account/session metadata, or any other record connected to the apprehended offence, even if it seems routine to do so as part of normal data-retention practice.
- Restrict access to the preserved material to only those directly handling the report, to protect the child's privacy and the integrity of the evidence.
- Suspend routine retention/deletion timelines (see our Privacy Policy, Section 6) for the specific records involved, until law enforcement confirms they are no longer needed. Pursuant to Section 8(7) of the Digital Personal Data Protection (DPDP) Act, 2023, this statutory evidence preservation duty legally overrides routine data erasure requests.
- Record what was preserved and when, so there is a clear internal chain of custody from the point the apprehension was first raised.
4.4 What Happens on Our Platform & Minor Identity Protection (JJ Act 2015 Sec 74)
Independent of the statutory report, Khypri AI immediately restricts the account(s) involved from further activity where a credible apprehension exists, and cooperates fully with the SJPU/police, including responding promptly to lawful requests for the preserved evidence in 4.3. In strict compliance with Section 74 of the Juvenile Justice (Care and Protection of Children) Act, 2015, Khypri AI strictly prohibits disclosing, publishing, or disseminating the name, address, school, photograph, or any identifying particulars of any minor involved in any safety investigation or legal proceedings, maintaining complete confidentiality of the child's identity at all times.
4.5 Good-Faith Reporting Is Protected
No Khypri AI employee, contractor, or partner will face retaliation, penalty, or adverse action for reporting a good-faith apprehension under this section, even if the apprehension later turns out to be mistaken. The risk we are guarding against is under-reporting, not over-reporting.
5. Safety Incident Response Process
Where a genuine safety incident is confirmed (for example, a student was exposed to harmful content despite the Safety Core), we follow this process:
- Step 1 — Containment: The specific gap that allowed the incident is identified and blocked as quickly as possible.
- Step 2 — Notification: Affected parents/guardians are notified.
- Step 3 — Root Cause Review: The relevant Safety Core layer(s) are reviewed and updated to close the gap.
- Step 4 — Record-Keeping: Confirmed incidents are logged internally to track our safety record over time.
6. Parental Controls
Parents/guardians can view their child's activity summary, request a copy of their child's data, and request correction or erasure at any time by contacting dpo@khypri.com (or Deputy DPO), consistent with the rights described in our Privacy Policy, Section 8 (Data Subject Request Process). Parents can also delete the account themselves with "Delete my account" in their profile settings. A dedicated self-serve parent dashboard with granular activity logs is planned and not yet available platform-wide; where it is unavailable, requests are handled manually by our team.
7. DPDP Act 2023 Alignment (India)
Khypri AI operates in line with India's Digital Personal Data Protection (DPDP) Act 2023 concerning children's personal data. Our operational boundaries include:
- Verifiable Parental Consent (VPC): Only a parent or guardian can register on Khypri AI; students cannot create an account directly. Registration captures the parent's date of birth and a self-declaration of adult status, followed by verification of the parent's email and mobile number. Only after the parent's own profile is complete can they add their child(ren) to the account — a student profile cannot exist without an already-verified parent account behind it. We ask the parent for explicit consent before processing any of their child's data; if that consent is not given, no processing of the child's data takes place.
- Unbiased Academic Support, Not Behavioral Profiling: Khypri AI is designed to give a transparent, unbiased analysis of test performance to help the student understand concepts and improve. We do not track student web browsing, location telemetry, or behavioral patterns outside their immediate academic workspace, and this data is never used for advertising or marketing. See our Privacy Policy for the full lawful-basis explanation.
- The Right to Correction & Erasure: Parents can view, modify, or request erasure of their child's information at any time. Verified erasure requests are processed with a full delete from production systems within 30 days.
- Data Retention: Student and parent data is retained only while the account is active. When the account is deleted it is deactivated at once and the data is erased after a short cooling-off period, and in any event within 30 days of a verified request, unless a longer retention is legally required.
- Grievance Officer: Data protection queries or grievances under the DPDP Act and Consumer Protection E-Commerce Rules 2020 can be directed to our Data Protection Officer & Statutory Grievance Officer, Debasmita Maharana (or Deputy Grievance Officer), at dpo@khypri.com.
8. GDPR Alignment (European Union)
For users in the EU/EEA, Khypri AI aligns its data handling with the General Data Protection Regulation:
- Lawful Basis for Processing: Student data is processed under parental consent and the legitimate interests of providing the service.
- Data Residency: European user data is hosted on EU-based cloud infrastructure.
- Cross-Border Transfer Safeguards: Standard Contractual Clauses (SCCs) governing transfers of EU data outside the EEA are incorporated within standard vendor data processing addenda.
- Data Subject Rights: Users and parents may request access, correction, restriction, or erasure of personal data by contacting our Data Protection Officer, Debasmita Maharana (or Deputy DPO), at dpo@khypri.com.
- Breach Notification: In the event of a data breach, we commit to notifying affected users and relevant supervisory authorities within 72 hours of discovery.
- End-to-End Encryption: Data is encrypted in transit using TLS 1.3 and encrypted at rest using AES-256.
9. COPPA Alignment (United States)
For users in the United States, Khypri AI aligns with the Children's Online Privacy Protection Act, which applies to children under 13:
- Minimal Data Collection: We collect no audio recordings, visual media, or precise geolocation data from children under 13.
- Verifiable Parental Consent: Because only a parent or guardian can register and verify their own identity before adding a child to the account, no child under 13 is ever able to provide their own consent or create an independent account.
- Parental Access & Deletion Rights: Parents may review the data collected from their child and request its deletion at any time by contacting our Data Protection Officer, Debasmita Maharana (or Deputy DPO), at dpo@khypri.com.
- No Conditioning on Excess Data: Participation in Khypri AI is never conditioned on a child disclosing more personal information than is reasonably necessary.
10. Cookies & Third-Party Processors
We use only strictly necessary cookies for login sessions and security — no third-party marketing or advertising cookies. Data is processed only by BigRock (hosting and email), OpenAI and Google (AI content generation and evaluation), Razorpay (payments), and Google Analytics (aggregate usage analytics on our public marketing website only, opt-in only). As we scale, we may add AWS or Google Cloud as additional or replacement hosting infrastructure. We never collect a child's legal name in the first place — only a parent-chosen display name is used to identify a child within the platform, including in any content sent to our AI processors. Sub-processors process data under binding enterprise data protection addenda and API privacy commitments. See our Privacy Policy for full detail.
11. Contact & Statutory Grievance Redressal
For any question, correction, deletion, or grievance request under the DPDP Act, Consumer Protection E-Commerce Rules 2020, GDPR, or COPPA, please contact our Data Protection Officer & Statutory Grievance Officer, Debasmita Maharana (or Deputy DPO / Grievance Escalation Team), at dpo@khypri.com. We acknowledge all grievances within 48 hours and resolve verified erasure requests within 30 days.