AI Safety at iTutor

Grounded in your materials

iTutor's tutoring is designed to answer from the learner's uploaded materials (textbooks, notes, slides) rather than from world knowledge alone. When the learner asks a question that can be answered from their material, the system retrieves the relevant passage and explains based on it. This significantly reduces hallucination on course-specific content compared to general-purpose chatbots.

For general-knowledge questions outside the uploaded material, iTutor will answer from broader training data, but will tell the student when it is doing so.

Honest about limitations

AI can make mistakes. iTutor presents answers as the AI's best understanding, not as authoritative truth. Important answers — especially in math, science, and high-stakes content — should be checked against the learner's primary source.

iTutor will say "I do not know" when the system has insufficient information to answer reliably. It will not fabricate a confident-sounding answer to fill the gap.

Content filtering

Both inputs and outputs are filtered against categories including sexual content, graphic violence, self-harm, hate speech, and illegal activity. The filtering is layered — at the prompt-classification stage, during generation, and again on the final output before it reaches the learner.

For accounts identified as belonging to minors, the filters are configured more strictly.

Age-appropriate language

iTutor adapts its vocabulary and reading level based on the learner's stated grade or the difficulty of the material. A primary-school student asking about photosynthesis gets a different explanation than a university biology student asking the same question.

Refusal posture

iTutor will not assist with academic dishonesty when the goal is clearly to bypass learning. It will not write entire essays, complete entire assessments, or produce text designed to be submitted as the student's own work without disclosure. It will help with brainstorming, outlining, drafting under guidance, and revising — and recommends students disclose AI assistance per their institution's policies.

iTutor refuses requests for content that could cause harm: instructions for weapons, illegal activity, content sexualizing minors, or detailed self-harm methods.

iTutor will not provide medical, legal, or financial advice as if from a licensed professional. When such topics come up, it explains the relevant concepts and recommends consulting a qualified human expert.

Voice safety

Voice tutoring sessions are processed in the moment to enable the conversation. Audio is not used to train models or shared with third parties for advertising. Voice transcripts are retained only as long as needed to provide the learning service and to support account recovery; learners can request deletion of voice data at any time.

Protections for younger learners

For learners under 13, iTutor strongly recommends institutional accounts (where the school is the data controller and parents have visibility) or supervised parent accounts. We do not knowingly create individual accounts for children under the minimum age set by their region's privacy law.

Institutional plans expose parental and teacher visibility into a learner's activity, including conversation summaries and progress.

Bias and fairness

AI systems can reflect biases in their training data. We actively work to reduce demonstrable harms: prompting frameworks that avoid stereotyping, periodic evaluations across different student demographics, and content review of the templates and curricula generated by our AI tools. We will not claim iTutor is free of bias — that would not be honest — but we treat bias as something to measure and reduce, not ignore.

Reporting a concern

If iTutor produces an answer you believe is unsafe, biased, factually wrong on something it should have known, or otherwise concerning, we want to hear about it. Use the in-app feedback button on any AI response, or email us at the address below.