For a university, AI literacy means the people who run admissions, registry, student finance and student services can tell what an agent did, why, on what evidence, and whether they are allowed to approve it. It is not a lecture on how models work. It is the ability to read a decision record and refuse a bad one.
Who in a university actually needs it?
Two populations need different things. Academic staff mostly use general AI tools for teaching and research; their literacy question is about academic integrity and assessment design. The agents that act, the ones that draft offer letters, update student records or chase unpaid fees, sit in professional services. That is where literacy has to be deep, because that is where a decision lands on a student.
Within professional services, three groups matter:
- Everyone who works alongside an agent: knows what the agent may do, what a held item is, and when to escalate.
- Reviewers and approvers: admissions officers, registry staff, finance officers who approve what the agent drafts. They need to read the record, find the deciding bound and decline with a reason.
- Owners: the heads of admissions, registry or student services who set the agent's limits and answer for its results.
What should each group be able to do?
| Group | What they must be able to do | What stays with them |
|---|---|---|
| All staff | Recognise the four verdicts: allowed, held, refused, withheld | Escalating anything that looks wrong |
| Reviewers | Read the record, check where each fact came from, approve, edit or decline with a reason | Every decision that affects a student's status, money or record |
| Owners | Read the agent's certified depth and the depth each action needs, understand what a lapsed agent certificate does, sign off limit changes | The limits themselves, and the fairness of outcomes |
The point specific to universities is provenance. A transcript from another institution, a reference letter or a fee waiver request often arrives by email. That is an external fact. A reviewer who cannot tell an external claim from a sealed entry in the student record system will approve things they should question. Literacy here is learning to ask where a fact came from before asking whether it looks right, which is the core of decision literacy.
The second specific point is fairness. Many universities are public bodies and face additional transparency and fairness duties. A reviewer approving an agent's draft rejection letter needs to know that "the agent recommended it" is not a reason a student can be given.
An example: the admissions agent
Take a mid-sized university that uses an agent to screen applications against published entry requirements. The agent reads each application at depth 1, drafts a conditional offer at depth 2, and holds anything outside the standard requirements for an admissions officer.
An application arrives with a foreign qualification the agent has no mapping for. The agent holds it and names the bound: the source for qualification equivalence does not cover that country. A literate officer sees that the hold is about a missing source, not a weak applicant. They check equivalence through the proper route, record the reason, and approve or edit the offer. An untrained officer might decline because "the system flagged it". That is the difference literacy makes, and it is visible in the record.
How do you evidence it?
Regulators and auditors will look for the same three things:
- A policy that says who must be trained, to what level, by role. The AI literacy policy template gives the fields.
- A training record per person: what they completed, what they passed, when it lapses.
- A link between training and authority: the people approving held items are the people who passed the reviewer level, and their approval rights pause if their certificate lapses.
For universities operating in the EU, the AI Act's Article 4 asks providers and deployers to take measures to ensure a sufficient level of AI literacy among staff dealing with AI systems on their behalf. GDPR's Article 22 limits decisions based solely on automated processing that have legal or similarly significant effects, which is why an admissions decision needs a person who can actually judge it. Prompt-writing courses do not meet either need on their own; prompt literacy is useful, but it is not oversight. The full map of levels is in the guide to AI literacy and certification for people who work with agents.
Where the Academy fits
Pryme Intelligence Academy splits the work the way a university needs it split. Foundations, about four hours and free, gives every member of professional services staff the shared vocabulary. The Operator track trains the admissions officers and registry staff who approve held work, and ends in a written exam and a two-week practicum of real review decisions. Certificates are issued through VerifyDoc.ai and can be checked at source, so the record of who is qualified is not a spreadsheet.
Questions people ask
Do academic staff need the same training as admissions staff?
No. Academics using AI tools for teaching need guidance on integrity and assessment. Staff who approve an agent's decisions about students need reviewer-level training with a test of real decisions.
Should students be told an agent was involved?
Students should be told when they interact with an AI system, and any decision about them should come with a reason a person can stand behind. The agent's draft is not that reason.
What is the first agent a university should train people for?
Pick a high-volume, low-harm task such as answering routine student service questions at depth 2, so staff learn to review drafts before any agent touches offers, fees or records.