Our principles
Human oversight
AI supports decisions; people make consequential ones. Outbound messages and record changes proposed by AI will require approval by default.
Transparency
Users can see when content is AI-generated and why a suggestion or flag was made.
Privacy
AI uses only data the user may access. Institutional data is not used to train public models.
Fairness
We exclude sensitive attributes from risk and prioritisation features and monitor outcomes across groups.
Reliability
We test AI features before release, track errors and provide ways to report problems.
Institutional control
Each AI feature is designed to be enabled, configured or disabled by the institution.
Controls planned for institutions
| Control | Purpose |
|---|---|
| Feature switches | Turn individual AI features on or off by role or department |
| Approval requirements | Decide which AI outputs require human sign-off |
| Signal configuration | Choose which data may inform early-warning or prioritisation |
| Audit logs | Review AI suggestions, approvals and resulting actions |
| Feedback | Report incorrect or inappropriate outputs for review |
What we will not build
We will not build features that make automated admission or disciplinary decisions about individuals, infer sensitive characteristics, or monitor students in ways they would not reasonably expect. If an institution’s requirements move toward these areas, we will say so plainly.
Frequently asked questions
Is institutional data used to train AI models?
Institutional data is not used to train public AI models.
Can an institution turn off AI features?
Yes, that is the plan. Each AI feature can be enabled, configured or disabled by role or department.
How are AI errors reported?
Users can flag incorrect or inappropriate outputs, which are reviewed and used to improve the feature.