Responsible AI

Purpose

This document sets out how Insight EDU uses AI responsibly. It describes our current approach and is honest about what has not yet been independently reviewed. It is Version 1.0 and will be extended as our Responsible AI programme matures.

Responsible AI & Ethics principles

• AI assists teachers; it never replaces professional judgement.
• We collect and send the minimum information needed, and redact obvious identifiers before AI processing.
• Feedback is designed to help students improve their own work — we do not rewrite student work for them.
• We use supportive, age-appropriate language and avoid labels that could harm a student's confidence.
• We are transparent about where AI is used and about its limitations.

AI risk register

We track the main risks of using AI in a school setting and the controls we apply:

• Incorrect assessment — Control: AI grades are suggestions only; an authorised teacher must review before any grade is official.
• Exposure of personal information — Control: identifiers minimised, obvious PII redacted, requests sent with retention disabled (store=false), student work not stored in AI files or vector stores.
• Inappropriate or harmful output — Control: structured prompts, supportive-language rules, and an issue-reporting path; harmful output is handled as an AI safety incident.
• Over-reliance on AI — Control: teacher oversight is built into the workflow and cannot be bypassed by students.
• Model or provider change altering behaviour — Control: model changes are recorded in the Model Change Register and reviewed.

AI safety testing

We test the AI features before and after significant changes. Current internal testing covers feedback quality, assessment behaviour, and that identifiers are excluded from requests. Independent, third-party AI safety and prompt-injection testing has not yet been completed and is tracked as an external action.

Bias and fairness review

We are committed to fair treatment across year levels and student groups. We calibrate feedback to the student's year level and use supportive language. A formal, independent bias and fairness audit has not yet been completed; until it is, we do not claim the system is free of bias. Concerns can be reported for review.

Human oversight

A qualified teacher remains responsible for every official assessment decision. Teachers can change any AI-suggested grade, view the curriculum evidence behind a suggestion, and choose what feedback is released to a student. Students cannot approve their own assessments.

AI incident handling

Suspected AI safety incidents (for example harmful output or a material assessment error) are logged, triaged and handled under our internal Incident Response Plan, including containment, investigation, any required notification, and follow-up actions.

Model change register

We record the AI models in use and material changes to them. The current assessment and feedback model is OpenAI gpt-5.6-terra. When a model or provider changes, we re-check safety, privacy and assessment behaviour before relying on it.

Reporting concerns

Responsible AI questions or concerns can be sent to insighteducationaus@gmail.com or raised through the in-app issue report.