A university's governing body needs enough AI agent literacy to protect decisions that shape students' lives. Governors should understand how far agents may act in admissions, assessment, student services and finance, how that authority is certified and expires, what the record can reconstruct when a student appeals, and which member of staff answers for each outcome.
Which university decisions should worry the governing body most?
Universities run many processes that look administrative and are not. An admissions decision, a mark, an academic misconduct finding or a decision on a fee refund can change a student's course. The EU AI Act treats some uses of AI in education, such as decisions on admission and the evaluation of learning outcomes, as high-risk, which brings human oversight obligations for institutions in its scope.
The governing body does not need to understand models. It needs to know which of these decisions agents touch, how far they go, and who signs.
What must governors understand?
- Depth (per agent), and the depth each action needs. 1 read, 2 draft for a person to send, 3 routine work within bounds, 4 money or hard-to-undo actions, 5 coordinating other agents. An admissions agent might read applications at depth 1 and draft offer letters at depth 2, and never decide.
- Earned authority. Depth comes from certification on the university's own cases, within the role's ceiling. The lower wins.
- Expiry. Certificates lapse after 90 days, and a run that clears no depth drops an agent to depth 1 the same day. Plan certification runs around clearing and results periods, not during them.
- The record. Every verdict is filed append-only. When a student appeals, the university can show what the agent drafted, from which evidence, and who decided.
- Provenance. A transcript issued by the awarding body is signed evidence; a personal statement is declared. An agent should weigh them differently, and reviewers should know why.
What stays with academics and staff?
| Work | Agent | Stays with a person |
|---|---|---|
| Student enquiries about deadlines and services | Answers within bounds at depth 3 | Anything about an individual case |
| Admissions files | Summarises at depth 1 to 2 | The admission decision |
| Marking support | Drafts feedback at depth 2 | The mark |
| Academic misconduct cases | Assembles evidence | The finding and any penalty |
| Fee refunds | Prepares and holds at depth 4, if certified | Releasing them |
How should student welfare be handled?
Students contact universities about money, housing and mental health, often through the same channels as routine questions. An agent answering a timetable query may receive a message that signals distress. Train agents and staff so that such messages go to a person immediately, with no queue. Governors should ask how this hand-over works, who receives it out of hours, and how it is recorded.
This is also where literacy pays off among frontline staff. A student services adviser who knows the agent is at depth 3 for routine answers, and that welfare is outside its scope, will pick up those conversations quickly.
What should the governing body ask for?
A short report each term covering agent uses and depth, failed certification runs, appeals involving agent drafts with one reconstructed from the record, and approvers' credentials. A readiness assessment in a university gives the baseline; a plan to roll out AI agent training across a university closes gaps. The senior leader running the deployment can be examined through a board review, as in Pryme Intelligence Academy's Architect track. A bank board and a contact centre board face related questions. The full model is in AI literacy and certification for people who work with agents.
Questions people ask
Can an agent mark student work?
It can draft feedback and suggest marks for an academic to review. The mark, and responsibility for it, stays with the academic.
Should students be told when an agent handles their enquiry?
Yes. Clear disclosure is good practice and, for institutions serving people in the EU, Article 50 of the EU AI Act requires people to be told when they are interacting with an AI system.
Which governors need the most training?
Members of the audit and risk committee and any committee overseeing academic standards, since they will review appeals and incidents.