School AI rules are a national story.

New York paused student-facing generative AI through eighth grade. Los Angeles restricted it on district devices for every grade. Those were From the Lab notes 001 and 002. They are not a coast-only story.

Outside New York and California, the country is mostly not running permanent student bans. The early 2023 ChatGPT network blocks in places like Seattle, Miami-Dade, and Orange County largely gave way to something else: approved-tool lists, teacher-permission regimes, and AI literacy frameworks. Full “no student GenAI forever” policies are rare. The sharper line is staff-first adoption with restricted or no direct student access — Chicago Public Schools is the clearest big-district example: staff use vetted tools; students do not currently get direct generative AI at school.

States are writing that into law. Tennessee already requires every local board to adopt an AI-use policy. Ohio set a July 1, 2026 deadline for the same. Maryland, Virginia, North Carolina, and Utah are phasing guidance, local policies, coordinators, and literacy standards through 2026–28. Oklahoma's SB 1734 adds human review before high-stakes use and a parent opt-out from student-facing AI without academic penalty. Florida has a proposed State Board rule that would flip the default the other way — parent opt-in, a public approved-tool list, and a bar on companion or social-emotional AI — but that rule is not adopted yet.

Districts that are adopting are doing it with guardrails, not open internet. Houston rolled ChatGPT Edu, Gemini, and Copilot to educators under data agreements. Denver lets students use named tools on school devices with teacher-set limits per assignment. Philadelphia restricts district GenAI work to approved Gemini (staff and grades 9–12) and Adobe Express/Firefly (K–12). Orange County, Florida, passed a formal classroom AI policy in July 2026: human-in-the-loop, district-approved platforms only, no AI as the final call on high-stakes decisions.

Guidance without a mandate is still spreading — Washington's human-centered OSPI guidance, Massachusetts' voluntary DESE materials, DC's September 2026 model policy booklet for LEAs. The arc since 2023 is clear enough to say out loud: ban, then guide, then adopt under rules. Treat “ban” as the historical chapter, not the national end state.

That is the news.

Tech Lab Take

NYC and LA made the headlines. The rest of the map is where the decade actually gets decided. If your district is in Tennessee, Ohio, Maryland, or Oklahoma, AI in schools is no longer optional paperwork — it is a compliance and judgment problem. If you are in Chicago, Seattle, or Philadelphia, the live question is who gets the tool and who only watches a teacher use it. If you are watching Florida's draft rule, the fight is consent: opt-out versus opt-in before a child ever opens a prompt.

The lab's stance does not change with the zip code. Pace the rollout. Put tools and knowledge in adults' hands before the product outruns the room. Measure what a year of restraint or a year of governed use actually does, then publish it. That is the same receipt we argued for on the coasts — and the same reason Pacing the Frontier asked for brakes that work at the frontier. Rushing a national stack of approved chatbots into classrooms without the prep digs potholes in Ohio and Oklahoma the same way it would in Queens or the Valley.

This is not anti-AI. It is refusing to treat “everywhere” as a reason to skip the evidence. Notes 001 and 002 were the trend. Note 003 is the map: school AI policy is a national story, and the adults who write it slowly — with human locks, consent defaults, and public results — are the ones leaving the next generation something stable to stand on.

— BK