How many states introduced AI-in-education bills in 2026?

FutureEd's tracker, updated July 13, 2026, counted 77 bills across 27 states, with 10 enacted. The bills range from graduation requirements to grading limits and literacy courses.

What should a school AI policy separate?

Student-facing generative AI, teacher productivity tools, and teacher-facing instructional intelligence are different uses. A policy that treats all three as one thing will ban too much or permit too much.

77 Bills, 27 States, One Deadline: What a School AI Policy Should Actually Decide

Q: What did Ohio require by July 1, 2026?

Ohio Revised Code 3301.24, enacted through House Bill 96, required every public school district, charter school, and STEM school to have an AI policy. The state released a model policy in January 2026.

Q: Does vendor review already measure instructional impact?

Not always. NYC's ERMA process today focuses on privacy and security. Instructional-impact review is a stated next step, not the current full gate. Policies should still ask which decision a tool improves.

On July 1, 2026, every public school district, charter school, and STEM school in Ohio had to have an artificial intelligence policy on the books.

The requirement comes from Ohio Revised Code 3301.24, enacted through House Bill 96. The state released a model policy in January 2026 covering acceptable use, privacy and FERPA compliance, vendor evaluation, academic integrity, a standing AI workgroup, and periodic review.

Ohio is not alone. FutureEd's legislative tracker (updated July 13, 2026) counts 77 AI-in-education bills across 27 states in 2026, with 10 enacted so far. Among them:

Alabama H.B. 329: all students must complete a computer science course that includes AI instruction to graduate.
Maryland S.B. 720: state guidance, district AI policies, designated coordinators, statewide professional development.
Oklahoma S.B. 1734: district AI policies by 2027–28; AI may not be the sole basis for grading or discipline; parents keep opt-out rights.
Utah H.B. 218: a required grade 7–8 digital skills course including AI literacy.
Virginia H.B. 1186: a pilot program and guidance on safe, ethical AI use.

The policy wave is here. The question is what these policies actually decide.

Most AI policies are about prevention

Read a typical district AI policy and you will find a list of rules: what students may not submit, what data vendors may not collect, what teachers may not delegate.

That is necessary. It is not sufficient.

A policy that only prevents harm leaves the most important question unanswered: what instructional decision should get better because this technology exists?

Without that answer, a school ends up compliant and directionless at the same time.

The three things a policy should separate

Most confusion in AI policy comes from treating three very different uses as one thing.

1. Student-facing generative AI. A student asks a system to write, solve, or explain. This is where cognitive-ownership risk lives, and where New York City and Norway drew hard lines this year.

2. Teacher productivity AI. A teacher drafts a worksheet or a parent email. Low risk, real time savings, limited instructional value.

3. Teacher-facing instructional intelligence. Evidence from assessments is turned into a specific diagnosis and a specific next teaching action. The student never touches the AI. The teacher decides.

A policy that treats all three as "AI" will either ban too much or permit too much. A policy that separates them can be strict where it should be and ambitious where it can be.

Vendor evaluation is the clause that changes behavior

The Ohio model policy includes vendor evaluation requirements. New York City's announcement pointed in a similar direction.

Be precise about what is in force today. NYCPS reviews tools through ERMA for data privacy and security. The district said it is building expanded evaluation capacity, including instructional effectiveness and equity impact. That instructional-impact review is a stated next step, not the current full gate.

Even so, the standard schools should ask vendors is the same: not "does it have AI?" but "which decision does it improve, and can you show me?"

Schools should ask for evidence in three forms:

What did the tool see that the teacher could not?
What action did it recommend?
Did the gap close, and how fast?

A gap the Ohio analysis flagged

One legal analysis of the Ohio model policy (KJK, June 12, 2026) noted that it is "largely silent on special education."

That matters. Students with disabilities and multilingual learners are the ones most exposed to AI that masks a gap instead of surfacing it. NYC's policy explicitly exempted assistive technology for these students. Districts adopting model language verbatim should decide, on purpose, how AI evidence feeds into IEP and intervention decisions.

What we would put in a policy

Our bias is clear, so we will state it. Omnix360 is teacher-facing instructional infrastructure. Students never interact with the AI. The loop runs behind instruction: Plan → Teach → Diagnose → Verify.

If we were writing a policy, we would add one sentence to every prevention clause:

"Any AI system used for instruction must make the teacher's next decision more specific, and must show evidence that it did."

That single standard rules out most of what is being sold to schools. It also makes room for what could actually help.

See teacher-facing AI and misconception radar.

The takeaway

Compliance deadlines produce policies. They do not produce clarity.

The states and districts that get this right will not be the ones with the longest list of prohibitions. They will be the ones that decided, in writing, which instructional decisions AI is allowed to make better.

See Teacher-Facing AI in Action.

Sources

KJK: Ohio's July 1, 2026 School AI Policy Deadline (June 12, 2026)
Ohio Department of Education and Workforce: AI Model Policy (January 2026)
FutureEd: 2026 State AI in Education Bills tracker (updated July 13, 2026)
New York City Mayor's Office: generative AI moratorium (September 2, 2026)