Class rules and guardrails
Set the help students receive and understand which restrictions are enforced by the workspace.
Start with the assignment#
Tell students what they may use AI for. A practice exercise, a draft critique, and an assessment can need different rules. Write the class instructions in terms a student and a model can follow.
For a practice problem, an instruction might be: "Ask the student what they have tried. Give one hint at a time. Ask them to explain the next step before continuing."
Set the learning policy#
An assigned teacher or school operator can save the class learning policy. It includes the learning mode, class instructions, and approved model choices. The server checks the caller's class access when the policy is read or changed.
School conversations use a class-bound policy. The bound policy endpoint provides the policy associated with a conversation.
Separate instructions from access controls#
Class instructions guide the model's response. Access controls determine whether a person can open the class, use a model, or read a conversation. A prompt asking the model to ignore instructions does not grant a new model permission or teacher role.
Model instructions alone cannot guarantee that every response is appropriate. Test the policy with realistic student questions and review the responses.
Review a change#
After saving a policy, check the saved values and start a new class conversation. Confirm that the available model choices and help level match the assignment. Existing conversations can retain their bound policy context; review that context when investigating an answer.
Report a response that misses the mark#
Save the conversation reference and explain which class instruction the answer failed to follow. Share it through your school's review process. Do not publish a student's private conversation as a support example.