The tutor follows written rules on every turn. An external AI model is off unless the institution turns it on and each student agrees. The tutor cannot give the final answer: it knows the step the student is on and the correct value, and it is built not to reveal either.
Hints, "why this step" and the next question come from the course content and from what the student has shown they can do. Grades never come from an AI model.
The institution turns it on per deployment; each student is asked first and can withdraw at any time. Without consent, the tutor follows its written rules alone.
A hint that contains the answer, or the value of the next step, is stopped before it reaches the student, whichever way it was produced.
Recorded as a setting alongside the provider's terms, and stated in the residency statement produced for your deployment.
Instructor of record and co-instructor by default; admins and coordinators off by default; teaching assistants never. The privacy officer reviews grants; the student sees every opening.
What the tutor is made of, what it may and may not do, and which settings govern the external AI model, in one document. Available on request.
The Abu Dhabi school regulator frames AI in education around six safeguards. This table says what the product does under each; it is a description, not a certification.
| Safeguard | What the product does |
|---|---|
| Data protection and student privacy | One database per institution in a UAE region; retention per data class; a privacy-officer seat; export, correction and deletion inside the product; the external AI model off unless the institution turns it on and the student agrees. |
| Cognitive agency | The tutor explains, hints and asks back inside the step. It cannot give the answer or the next step's value, so the student does the work and the product records what they could do. |
| Academic integrity | Every student gets their own numbers in each question, and the product checks the answer itself; no hint can hand over the answer; checkpoints are announced with their stakes shown. |
| Human oversight | Grades follow the coordinator's policy and never come from an AI model. Instructors decide every intervention. Who may open a transcript is granted, reviewed by the privacy officer and visible to the student. |
| Content safety and cultural alignment | The tutor draws only on course content written from scratch and reviewed by a named faculty member before a pilot, in English and Arabic. An external AI model enters only when the institution turns it on and the student agrees. |
| Algorithmic fairness | The same rules and the same kind of questions for every student, in both languages. Grades never come from an AI model. In a pilot, no student in a comparison section is denied help, and that is agreed in writing before the term. |
Ask it for the answer and watch it refuse, in English or Arabic.