Policy & governance

“Students are already using AI — and we have no policy.”

An AUP update can't govern AI. You need policy that covers automated decisions, vendor data practices, and staff use — not just ChatGPT in the classroom.

The clear-eyed read

What's actually going on

You are not behind because you are careless; you are behind because the tools moved faster than any governance cycle can. 81% of teachers report their district has no formal AI policy — while a large majority of students already use generative AI. The gap between practice and policy is now the norm, not the exception.

The instinct is to bolt a paragraph onto the acceptable-use policy. That instinct is wrong, because AI in a district is not one thing. It is a classroom tool, yes — but it is also an automated decision-maker in HR screening and scheduling, a data pipeline inside vendor products, and a silent feature update in software you bought three years ago. If your policy only addresses students and chatbots, it is governing a fraction of your actual AI use.

Michigan's MDE guidance (May 2026) is a meaningful first step, aligned to the MICIP continuous-improvement cycle. But it does not provide board-adoptable policy language, an enforceable prohibited-uses list, a risk-tiering rubric, or an appeal process. Treat state guidance as a floor, not a ceiling — and build the rest deliberately. Vague guidance is not policy.

First moves

30 / 90 / 365 days

Sequenced, not simultaneous. The 30-day moves cost little and buy you room; the year is where the change becomes structure.

First 30 days

  • Stand up a small AI workgroup with a named owner — not a committee of everyone.
  • Inventory actual AI use: tools staff use, AI features inside existing vendor products, and what students report using.
  • Adopt an interim staff rule set (approved tools, data rules) so guidance exists while policy is drafted.

By 90 days

  • Draft policy against the twelve components of a complete district AI policy — from purpose and definitions through procurement and review.
  • Run the draft past legal counsel and your board policy committee; borrow tested language from strong model policies rather than starting blank.
  • Set the risk-tiering rubric: which uses are low-risk, which need review, which are prohibited.

Within the year

  • Board adoption, public communication in plain language, and a standing review cycle — policy that never gets revisited is policy that stops working.
  • Fold AI review into procurement so every new tool and every vendor feature change passes through the same gate.

Carry the work

The tools that carry this