# Schools | AI Wars of 2026

> In research: the teacher as the deployment surface. Source-backed analysis from AI Wars of 2026.

Canonical URL: https://ai-wars.correax.com/chapter/schools

## Schools

### In research

> **In school systems the deployment surface is the teacher, not the student.**
> This chapter is still in research. Its topics show what the next evidence
> increment must answer.

School deployments are unusual among the settings this book examines. The buyer
is a ministry or district, the user is a teacher, and the beneficiary is a child
who never appears in the procurement decision. Each of those separations creates
a measurement problem.

The questions this chapter has to answer:

- **Teacher capacity.** Does an assistant reduce the administrative load
  teachers describe, or add a new tool to manage on top of it?
- **Safeguarding and age-appropriate use.** What controls govern what a system
  may do when the person on the other side is a child, and what evidence exists
  that the controls operate?
- **Procurement and scale.** Public-sector agreements can move very large
  numbers of seats at once. What follows a provisioning commitment, and how long
  does the gap between provisioned and used run?
- **Measurement.** When the intended outcome is learning, what would show that
  teaching changed? Teacher satisfaction and learning gain are both worth
  knowing and are not substitutes for one another.

The strongest evidence currently held is a vendor-published account of a
national partnership that pairs a very large access commitment with a small,
dated pilot reporting band-level reading movement. The two numbers in that pair
are useful precisely because they are so far apart. [@microsoft-deped-copilot-teachers-2026]

A second vendor account describes U.S. school-district partnerships with stated
access, training, and privacy arrangements and no usage or outcome measure.
[@openai-chatgpt-teachers-districts-2026]

### What will make this chapter publishable

Independent evaluation with a comparison group, from an inspectorate, ministry,
or national evaluation body rather than a vendor. Until then the chapter can
report what vendors state, name each design limit, and decline to aggregate
separate pilots into a trend.

The chapter must make no claim about children's outcomes beyond what a cited
study establishes, with its population, period, and design stated alongside it.