Child Safety

Last updated: 21 May 2026 · Version 2.0

AL1 Tutor is built for children from Primary 1 through Junior College. Their safety — physical, emotional, and digital — is our highest priority. This page explains how we protect children using the service.

1. Our Commitment to Child Safety

We build AL1 Tutor knowing that children — sometimes very young children — will be reading what our AI writes and may treat it as authoritative. We accept the responsibility this places on us. Our approach is shaped by:

• the requirements of Singapore's Personal Data Protection Act 2012 and the PDPC's Advisory Guidelines on the PDPA for Children's Personal Data in the Digital Environment;
• guidance from the IMDA-supported AI Verify Foundation, including the Project Moonshot framework for AI red-teaming;
• the practical reality that AL1 Tutor sometimes encounters children expressing distress, asking about sensitive topics, or pushing the boundaries of what a tutor can help with. We have a duty to respond responsibly.

2. Age Requirements & Account Structure

AL1 Tutor accounts are owned by adults — parents, legal guardians, or authorised educators. Account holders must be at least 18 years of age. Children do not create their own AL1 Tutor accounts.

Within each adult-owned family account, the parent may create up to six (6) student profiles for the children in their household. Each profile is identified only by a name or nickname (chosen by the parent), an educational level, and the subjects the child is studying. We do not collect:

• date of birth, NRIC, FIN, or any government-issued identification;
• school name, class, or teacher's name;
• profile photos, avatars, or other personal images;
• any biometric data beyond what is strictly necessary for optional features (e.g. oral practice speech-to-text, which is not retained beyond the session).

This design means that even in the unlikely event of unauthorised access to our systems, the data exposed for a child is limited to their academic content — not their identity in the offline world.

3. Content Moderation

Several layers of moderation operate on every interaction:

On the way in (what your child sends)

• Image uploads (spelling list photos, worksheet photos) are screened by an automated safety model before being passed to any AI tutor. Inappropriate content is blocked at this stage.
• Chat messages are filtered for prompt-injection attempts and content that would attempt to manipulate the AI into producing unsafe output.
• Voice recordings (where applicable) are converted to text and the text is moderated.

On the way out (what the AI says)

• AI responses are constrained by carefully designed system prompts that establish AL1 Tutor's role as a tutor: helpful, accurate, age-appropriate, and aware that it is speaking with a minor.
• Responses on sensitive topics (violence, sexual content, self-harm, substance use) are filtered. The AI is configured to acknowledge a child's feelings without engaging in detail and to redirect them to trusted adults.
• Composition feedback and model essays are generated at a register appropriate to the child's level — we do not, for example, suggest mature vocabulary to a Primary 4 student.

4. AI Safety Approach

AL1 Tutor uses third-party large language models (currently Claude, Gemini, and others) under enterprise API terms. These providers maintain their own safety training and content policies. We layer additional safeguards on top:

• Retrieval-augmented generation: for academic content, the AI is grounded against curated MOE syllabus material and verified question banks, reducing the risk of fabricated answers.
• Auditor models: a second AI model independently audits the primary tutor's responses for serious factual errors before they are shown to the child.
• Red-team testing: we test AL1 Tutor against known prompt-injection, jailbreak, and content-extraction techniques. We plan to integrate with the IMDA AI Verify Foundation's Project Moonshot testing framework as it matures.
• Confidence signalling: AL1 Tutor signals to the user when a response has been verified against source material (a "verified" badge) versus when it has been generated without external grounding.
• Human review of edge cases: conversations flagged by the system as anomalous are reviewed (with personally identifying information redacted) so that we can continuously improve the safeguards.

5. What We Do Not Do

AL1 Tutor does not:

• show any advertising to anyone, ever — not to parents, not to children, anywhere on the platform;
• permit children to communicate with each other through the platform (no chat rooms, no forums, no leaderboards with usernames, no peer messaging);
• permit children to communicate with strangers in any form;
• use children's data to profile them for any commercial purpose other than delivering the service;
• sell, share, or rent your data or your child's data to third parties for marketing;
• use children's User Content to train external AI models;
• use third-party tracking pixels or advertising cookies;
• encourage children to invite friends in exchange for in-product rewards (referral codes are issued to parents, not children).

6. Handling Emotional or Safety Concerns

Children occasionally tell tutors things that are not about homework. AL1 Tutor's AI is configured to handle this with care:

• If a child expresses distress, sadness, anxiety, or signs of being overwhelmed, the AI acknowledges the feeling without psychoanalysing, and gently suggests they speak to a parent, guardian, school counsellor, or trusted adult.
• If a child indicates self-harm, abuse, or imminent danger, the AI will not engage with the specifics of the situation. Instead, it will recommend the child immediately speak to a trusted adult and, where appropriate, provide contact information for relevant Singapore helplines (see Section 9).
• The AI never asks a child to keep secrets from adults, never elicits private personal information for non-academic purposes, and never positions itself as a substitute for human connection.
• For more sensitive situations, AL1 Tutor's chats may be flagged for review by our safety team (with redaction of personally identifying information) so that we may improve our handling of similar situations in future.

If you have set up parental notification (see Section 7), repeated flags of this type will result in an alert to you.

7. Parental Controls

As the parent or guardian, you have access to:

• full visibility into every conversation your child has had with AL1 Tutor, accessible from the family dashboard;
• per-student progress and mastery reports;
• the ability to delete any individual chat, spelling list, or composition;
• the ability to delete an entire student profile and all associated data at any time;
• weekly summary reports via WhatsApp (where you have opted in), giving you a digest of what each child has worked on;
• the ability to opt your account out of analytic use of de-identified content, via the Privacy settings.

We recommend that parents periodically review their children's AL1 Tutor activity. This isn't because we expect problems — it is part of the same healthy supervision that any digital tool with a child warrants.

8. Reporting Concerns

If you encounter content, behaviour, or output from AL1 Tutor that concerns you — whether on safety, accuracy, or child welfare grounds — please report it immediately:

Where a concern relates to alleged harm to a child by another person (not by the AI), we will work with you on the appropriate Singapore authorities to contact.

9. Singapore Resources

If you or your child needs to speak with a trained human counsellor outside AL1 Tutor, these Singapore services are available:

• Samaritans of Singapore (SOS) — emotional support, 24-hour hotline: 1767. Website: sos.org.sg
• Tinkle Friend (for primary school children) — Singapore Children's Society helpline: 1800-2744-788. Website: tinklefriend.sg
• National Care Hotline — 1800-202-6868 (operated by MSF).
• School counsellors — every MOE school has counsellors trained to support students. Encourage your child to use this resource where appropriate.

AL1 Tutor is a tutor. It is not a counsellor, therapist, or crisis line. We will always defer to human professionals when a child needs support beyond academics.