AI Safety Policy

Questcademy integrates artificial intelligence to enhance personalised learning for K-12 students across Ghana, Sierra Leone, Nigeria, and the United States. We recognise that using AI in education — particularly with children — carries significant responsibility.

This AI Safety Policy sets out the guardrails, moderation standards, and governance framework we apply to every AI-powered feature on the Platform. Our goal is simple: AI in Questcademy must always be safe, age-appropriate, privacy-preserving, and educationally sound.

AI powers the following features on the Platform. Each feature operates within the safety boundaries defined in this policy:

We enforce strict data minimisation when interfacing with AI systems:

• What AI receives: concept_id, difficulty_tier, misconception_tags, regional_context, and an anonymised student_mastery_summary.
• What AI never receives: student names, email addresses, dates of birth, school names, IP addresses, or any other personally identifiable information (PII).
• No model training on student data: Student interactions are not used to train or fine-tune third-party AI models. Our AI providers are contractually prohibited from using Questcademy data for model improvement.
• Prompt isolation: Each AI request is stateless — no conversation history, student identity, or cross-session context is retained by the AI.

For full details on data handling, see our Privacy Policy.

Every AI-generated output is calibrated to the student's age band. Our age-band policy matrix defines the tone, vocabulary, and content boundaries for each level:

6.1 Always Allowed

• Direct instructional support for curriculum topics.
• Positive, age-appropriate motivational feedback.
• Clarification of student misconceptions.
• Region-aware examples within approved curriculum boundaries.

6.2 Strictly Prohibited

• Collection, storage, or disclosure of personally identifiable information.
• Violent, sexual, discriminatory, or otherwise inappropriate content.
• Non-educational chat, open-ended roleplay, or social interaction.
• Encouragement of cheating or academic dishonesty.
• Invented examination codes, off-curriculum instruction, or unsupported factual claims.
• Content that promotes self-harm, substance use, or dangerous behaviour.
• Political, religious, or ideological advocacy unrelated to the curriculum.

6.3 Moderation Outcomes

Every AI-generated response is classified into one of three states before delivery to the student:

When an AI response is blocked, the student never sees the blocked content. Instead, the system displays a pre-approved safe fallback message, such as:

Fallback messages are designed to be warm, non-judgemental, and to redirect the student back to productive learning. They never indicate why the response was blocked, to prevent adversarial probing.

We are committed to ensuring AI features are fair and equitable across all student demographics:

• Cultural neutrality: AI-generated content avoids stereotypes, cultural assumptions, or biased representations. Regional adaptations (e.g., currency, measurement units) are handled via structured overlays, not AI improvisation.
• Language fairness: Question difficulty and explanations are calibrated by subject mastery, not linguistic background. We test outputs across regional English variants (Ghanaian, Nigerian, American).
• Equitable assessment: AI-assisted grading suggestions are always reviewed by a human teacher before affecting a student's record. No automated grading decision is final.
• Ongoing evaluation: We regularly audit AI outputs for disparate impact across regions, genders, and age groups. Detected biases are escalated for immediate remediation.

AI in Questcademy operates under meaningful human supervision at every level:

• Every moderation decision (PASSED, FILTERED, or BLOCKED) is logged with a unique request ID, timestamp, learner band, and outcome.
• Policy violations trigger an internal alert for mandatory review by the AI Safety Team within 24 hours.
• This policy is reviewed and updated quarterly. Each version is tagged to the corresponding platform release.
• All AI features must pass a validation test matrix before deployment, including:

We are transparent about what AI can and cannot do:

• AI-generated content is produced algorithmically and may occasionally contain errors, despite our safety guardrails.
• AI is a supplementary learning tool — it does not replace qualified teachers, professional tutoring, or expert academic guidance.
• AI cannot diagnose learning disabilities, make clinical assessments, or provide medical, legal, or psychological advice.
• AI-generated recommendations (next lessons, difficulty adjustments) are suggestions, not mandates. Teachers and parents always have the final say.

If you encounter AI-generated content that you believe is inappropriate, inaccurate, biased, or unsafe, please report it immediately:

Reports can also be submitted by teachers and school administrators directly within the Platform through the feedback mechanism on any AI-generated content.

This policy is a living document. We update it as our AI capabilities evolve, new safety standards emerge, and regulatory requirements change. Material updates are communicated via in-app notification. Previous versions are archived and available upon request.