AI Use & Limitations

1. What the AI Mentor is

The AI Mentor is a Socratic-style study aid, gated to the specific question a student is currently working on. It can break a problem into smaller steps, ask guiding questions, explain the underlying concept, rephrase an explanation, or give a further example — with the deliberate goal of helping a student reason toward an answer, rather than simply supplying one.

2. What the AI Mentor is not

• It is not a qualified teacher, tutor, or education professional.
• It is not a general-purpose chatbot — it is deliberately restricted to the current question, and the Service actively redirects messages that try to use it for anything else.
• It is not a source of guaranteed-correct information — see section 5.
• It is not a counselling, mental health, or crisis service — see our Child Safety / Online Safety policy for what happens if it detects a possible safeguarding concern, and for where to get real support.

3. How it works

When a student asks for help, the current question, their message, and (where relevant) an image associated with the question are sent to an AI model provided by Anthropic, accessed through the Vercel AI Gateway. The model is instructed, via a fixed system prompt specific to the current question, to act as a Socratic tutor and not to state the correct answer directly. Before a response reaches the student, a separate, non-AI check scans it for the literal expected answer and substitutes a safe fallback response if it finds one — this is a deterministic backstop, not something the AI model decides for itself.

Two further automated checks run before a message reaches the AI model at all: one detects off-topic or prompt-injection attempts (see our Acceptable Use Policy) and one detects possible safeguarding disclosures (see our Child Safety / Online Safety policy). A conversation is capped at a fixed number of messages, and requests are rate-limited per user.

4. How your information is handled

We configure our AI provider integration to opt out of using conversation content to train the provider’s models. We do not store the content of an ordinary AI Mentor conversation in our database — it is processed for that single request and not kept, except where a message is automatically flagged as a possible safeguarding concern, in which case it is logged as described in our Child Safety / Online Safety policy. We do record that a student used the AI Mentor on a given question (so a teacher can see help-seeking patterns), but not what was said. Full detail is in our Privacy Policy.

5. Known limitations

You should be aware that:

• AI-generated explanations can be wrong. Like any AI language model, the AI Mentor can produce an explanation that is incomplete, oversimplified, or factually incorrect, even for a question with a verified correct answer in our system.
• The answer-leak guard is a heuristic, not a guarantee. It reliably catches the correct answer stated in its literal form, but it is not designed to catch every mathematically or logically equivalent way of stating that answer.
• The off-topic and safeguarding checks are pattern-based. They can miss something they were not designed to recognise, or occasionally flag something they shouldn’t. They are not a substitute for human judgement or supervision.
• Rate limiting is not centrally enforced across every server. It is a practical first line of defence against abuse, appropriate to Mentor’s current scale, rather than a guaranteed hard limit.

6. Human oversight

Teachers do not see the content of a student’s ordinary AI Mentor conversations — only that help was used on a particular question, and (only where a message was flagged as a possible safeguarding concern) the flagged message itself, via the alert described in our Child Safety / Online Safety policy.

7. Reporting an AI safety issue

If the AI Mentor produces a response that is harmful, clearly wrong in a way that matters, or that you believe should have been blocked, tell your teacher and contact us at support@learnwithmentor.com.au with the question and, if possible, what the AI Mentor said.