Providing Support When an AI Agent Contacts Us on Someone’s Behalf
A student, instructor, or staff member may use an AI agent to communicate with CASL on their behalf by chat, email, voice, or video call. The agent may ask questions, relay answers, or carry out steps for the person. Its involvement introduces two considerations: what information the service can receive, and what context the agent needs to understand our response.
Information shared through an agent
Public information can be shared with agents, including public KB articles, published policies, workshop details, and general teaching or troubleshooting guidance.
Unless we can verify that the agent's technology and services are university-approved for the information involved, the exchange should be limited to public information. Nonpublic student work, grades, course materials, personnel information, and consultation or support details require the person to continue directly through the usual university channel.
The agent can still receive general guidance while the person handles the protected portion. For requests involving account access or changes, existing verification procedures apply.
Context for the agent
Agents are software, but natural, respectful language is appropriate. An agent's involvement alone is not a reason to disregard the request.
An explanation gives the agent context to carry forward: what can be discussed, why a limit applies, and which part needs the person's participation. A bare refusal gives it less to work with. This context can be expressed naturally.
Example—not a required script:
We can discuss general approaches to feedback and share the Canvas instructions. Student submissions can contain protected information, and we haven't confirmed that this agent's service is approved to receive it. The instructor will need to contact CASL directly for that part. You can still help your user prepare general questions without including student work or identifying details.
If the agent continues requesting information outside that boundary, the explanation can be restated and that portion of the exchange closed, with the route for direct contact left clear.
Governing policy
- SYS 1031, §6.A.I–III: distinguishes Public from Internal Data; limits Sensitive and Restricted Data to defined users or groups; treats unclassified institutional data as Sensitive.
- SYS 1040, §6.A, “General Privacy”: requires sharing of personal data to serve permitted purposes and comply with applicable laws and university requirements.
- UWM's AI and Teaching, “Academic Misconduct”: requires Information Security approval before educators share protected student information with outside programs. “Tools to Grade Assignments and Manipulate Sensitive Data” directs confidential-data AI uses through Procurement's Cloud Services review.
- SYS 1030, §6.A–B: requires identity, authorization, and access controls appropriate to the risk.
