The Ledger · Brief No. 002 · 2026-08-01

Every state regulates whether a student used AI. None asks whether the AI works.

We read all 51 state-level K-12 artificial intelligence policies in force across 40 jurisdictions and classified what each one actually governs. Start with your own state.

What the 51 policies actually govern

Each policy is classified by what it addresses. A single policy usually covers several things, so these do not sum to 51.

Acceptable use
44
Student data privacy
42
AI literacy
36
Bias and equity
32
Teacher professional development
32
Academic integrity (did a student use AI)
24
Procurement process
10
Efficacy or vendor review (does the product work)
10
Deepfakes
1

The finding, stated so you can disagree with it

24 state policies govern whether a student used AI dishonestly. 10 touch whether the product works at all, 6 of them bindingly, and not one requires a company to show that its AI improves learning before a district may put it in front of children.

Every one of those six is printed in full below, with a link to the statute, so you can read them and decide whether that reading is fair. If you think one of them does impose an efficacy requirement, we would like to be told, and it is covered by our standing correction offer.

The six binding statutes, in full

These are every binding state law that touches efficacy or vendor review. Read what each actually requires. Three regulate how teachers may use AI, one regulates privacy, one directs a department to build evaluation tools for districts, and one commissions a study.

StateLawWhat it requiresCheck
IdahoGenerative AI in Education Framework Act
S 1227 (2026), Session Law Chapter 71		Directs the Idaho Department of Education to develop a statewide generative-AI framework covering privacy, procurement safeguards, transparency, academic integrity, AI-literacy standards and PD; requires districts and charters to adopt aligned policies; sets procurement requirements and requires AI vendors to disclose their use of AI and provide data-protection assurances; prohibits AI from replacsource
MarylandArtificial Intelligence Ready Schools Act
SB 720 (2026), Chapter 634		Requires MSDE to provide AI guidance to local school systems via an online platform and to develop guidelines, best practices and EVALUATIVE TOOLS to assist local school systems in the evaluation of artificial intelligence tools selected for use; establishes the Maryland AI Education Collaborative; districts get 120 days from guidance issuance to adopt aligned policies.source
OklahomaOklahoma Responsible Technology in Schools Act
SB 1734 (2026)		Requires OSDE to develop AI and emerging-technology guidance and requires district boards to adopt conforming AI policies; mandates human-in-the-loop review of AI outputs used in instruction, feedback, assessment or decision-making; bars AI as the primary basis for grading, discipline, placement, promotion or retention; requires parent notification with opt-out and annual disclosure of AI tools ansource
UtahClassroom Technology Amendments - Model artificial intelligence use policy
HB 273 (2026), Laws of Utah 2026 ch. 120; Utah Code 53G-7-1401 to 1403		Requires the Utah State Board of Education to publish and maintain a model AI use policy and requires every LEA to produce, adopt, submit, train on, enforce and biennially review a conforming policy. The model policy must prohibit educators from using generative AI to independently grade student work or issue high-stakes determinations, require LEA approval of any student-facing AI tool, require wsource
UtahSoftware in Education Amendments (Software in public schools study)
SB 267 (2026), Laws of Utah 2026 ch. 161; Utah Code 53E-3-527		Requires the Utah State Board of Education to STUDY the use of software and digital services in public schools, including the academic effectiveness of educational software (whether it supports active learning, skill development and measurable academic progress), addictive design features, student data practices, and best practices for evaluating educational value, and to publish guidance to everysource
VirginiaGuidance and policies on the use of artificial intelligence systems in public elementary and secondary schools
HB 1186 / SB 394 (2026), Acts of Assembly 2026 cc. 937, 943; Va. Code 22.1-20.2:1		Requires the Virginia Department of Education to establish guidance on safe AI use in schools addressing student data privacy under FERPA and COPPA, teacher training, AI system transparency, data privacy agreements PROHIBITING external model training on student data, bias prevention and equitable access, and alignment with the Standards of Learning; requires every local school board to implement csource

And eleven jurisdictions have nothing at all

No state-level K-12 AI policy was identified in these, as of 2026-08-01. Two of them run among the largest school systems in the country.

Arkansas · Arizona · Colorado · District of Columbia · Florida · Iowa · Kansas · Nebraska · New Hampshire · New York · South Dakota

An absence is the hardest thing to establish, and these are the rows we would most like to be wrong about. If your state belongs here and has a policy we missed, that is the most useful correction anyone could send us.

Where this comes from, and what is soft about it

Source
State statutes, department of education guidance, state board policies and executive orders, each linked individually in the lookup above. 50 of 51 policies carry a direct source URL.
Retrieved
PILLAR Vertical Intelligence state AI policy registry, snapshot of 2026-08-01. This page re-derives every figure from that snapshot at build time and refuses to build if any of them moves.
The classification is ours
The categories are PILLAR's own coding, not the states' self-description. No statute labels itself as covering "academic integrity" or "efficacy review". We read each policy and tagged it. That is a judgement, it is the load-bearing step in this brief, and the source link is published for every row precisely so you can re-code it and disagree.
Confidence
34 of 51 policies are marked verified against primary text. 17 are marked needs-review, meaning the record exists but the primary source has not been independently re-read. Five of the six binding statutes below are verified; Idaho is not.
Known limit
This is state-level policy only. It says nothing about what individual districts require in their own procurement, and a district can impose an efficacy standard its state does not. It also captures policy in force, not bills in progress.
The word "none"
That claim is a reading of six statutes, not a field in the data. The build fails if the membership of those six changes, so the sentence cannot outlive the evidence behind it without someone noticing.
Think this is wrong? Read the six laws linked above. If one of them requires a vendor to demonstrate that a product improves learning, we will publish the correction with your name on it. The standing bounty and the correction log.

Take it and use it

The number

24 of 51 state K-12 AI policies govern whether a student used AI dishonestly. Zero require the company whose AI is teaching that student to show that it works.

The reversal

We built an entire regulatory apparatus for AI in schools, and pointed almost all of it at the children rather than at the vendors.

What has to travel with it

State-level policy only, as of 2026-08-01. The classification of what each policy covers is PILLAR's own coding of the text, not the states' own labels. 17 of 51 records are marked needs-review. Eleven jurisdictions have no policy identified, and an absence is the hardest finding to establish.

carry-kit.json and policies.json hold all of it in machine-readable form.