What we talk about when we talk about AI policy

Ask ten people at a university what the AI policy says and you will get ten answers, all of them partly right. That is not because the policy is badly written. It is because the word "policy" is being asked to carry three jobs at once.

The three documents inside one

The first is a compliance statement: what staff and students must not do, expressed in language that will hold up if it is ever tested. The second is a teaching philosophy: what we think learning is for, and which uses of a language model are consistent with that. The third is a procurement rulebook: which tools are approved, where data goes, who signs off.

These three have different audiences and different half-lives. The compliance statement changes rarely. The teaching philosophy should be argued about openly and often. The procurement rulebook goes stale every few months.

When a single document tries to do all three, the fastest-moving part drags the slowest-moving part with it — and nobody trusts any of it.

A more useful split

• Publish the compliance statement once and reference it everywhere.
• Let faculties own the teaching position, with a shared vocabulary rather than a shared rule.
• Keep the approved-tools list somewhere it can be updated without a committee cycle.

None of this is glamorous. It is, however, the difference between a policy people can follow and a policy people quietly ignore.