Between the 2023-24 and 2024-25 school years, the share of K-12 teachers using generative AI for their work doubled, from 25 percent to 53 percent, according to RAND's 2025 survey panels. Over the same stretch, the Walton Family Foundation and Gallup found that six in ten teachers used AI during the school year, and that the roughly one third who use it weekly save an average of 5.9 hours per week, which works out to about six weeks of reclaimed time across a school year. Now place one more number beside those: in that same RAND research, only 45 percent of principals reported having any school or district policy or guidance on AI at all.
RAND did not bury the conclusion. The 2025 report is titled "AI Use in Schools Is Quickly Increasing but Guidance Lags Behind." That sentence is the state of the profession in mid-2026. Adoption is a sprint. Guidance is a committee. And the teachers standing in front of students every day are being asked to close the distance between the two on their own.
I have spent the past three years facilitating AI professional development for educators across all 50 states, and I have read a great deal of the state guidance that now exists. This article is my attempt to answer the two questions teachers ask me most often. First, why does official guidance still trail so far behind what is already happening in classrooms? Second, what should I actually do while I wait? The first answer is structural, and understanding it will make you more patient with your state and less patient with excuses. The second answer does not require waiting at all.
Where state guidance actually stands in mid-2026
The most widely used count comes from the tracker maintained by AI for Education, which lists 34 states plus Puerto Rico with official K-12 AI guidance issued by a state education agency. Other 2026 tallies put the number anywhere from 33 to 35, depending on what qualifies as guidance. Some states have a comprehensive framework. Some have a memo. At least one, Tennessee, took a different route entirely, passing a law that requires every district to adopt its own AI policy rather than issuing state-level guidance. New Jersey, as of the most recent tracker updates, has curated resources but no official state guidance document.
So the honest summary is this: roughly two-thirds of states have something official on paper, and the remaining third have either nothing or a placeholder. Two years ago the count was close to zero, and I want to be fair about that. By the standards of education policy, which normally moves in decade-long arcs, this is fast. By the standards of the technology, it is geologic.
The more recent releases show the field maturing. Vermont's framework, released in January 2026, runs about 50 pages and is organized around human-centered implementation rather than a list of prohibitions. Georgia published its framework for ethical, effective, and secure use in January 2025. New Mexico developed its May 2025 guidance in partnership with university researchers. Montana built a biannual review cycle directly into its guidelines, which quietly concedes the central problem: any AI guidance document is partially obsolete the day it is published, and the states that admit this in advance are the ones taking the work seriously.
It also matters what these documents are, and what they are not. Nearly all of them are advisory. They recommend, they caution, they encourage. Very few bind a district to anything, and almost none arrive with funding attached. Read a stack of them side by side, as I have, and a pattern emerges: the sections on risk are long and confident, the sections on data privacy are careful and lawyered, and the sections on what a teacher should actually do differently on Monday are the shortest in the document, when they exist at all. That is not an accident. Risk is easy to write about because it requires no implementation. Practice is hard to write about because it requires knowing classrooms, and most guidance committees are several layers removed from one.
The legislative picture is moving even faster than the guidance picture. An analysis published by ExcelinEd in May 2026 counted eight states with enacted K-12 AI laws, more than 20 states with active bills, and over 1,500 AI-related bills introduced nationally in 2026 alone. Utah has enacted AI literacy requirements at the middle school level. Proposals in Maryland and Virginia would direct professional development and state guidance on teacher training. Bills in Oregon and Washington target the design of AI tools used by minors. Whatever your state's guidance says today, the legal floor underneath it is being repoured in real time.
Why the lag is structural, not a scandal
It is tempting to read the gap between classroom reality and state guidance as negligence. Having watched the process up close, I think that is mostly wrong, and the real explanation matters because it predicts what will and will not get better.
The first reason is that policy cycles and technology cycles run on different clocks. A state guidance document typically takes 12 to 18 months to move from task force to publication: convene the committee, gather stakeholder input, draft, review for legal exposure, revise, approve, release. Eighteen months ago, the dominant classroom concern was a chatbot that answers questions. Today's concern is agentic systems that complete multi-step tasks, which changes assignment design and academic integrity in ways a 2024 committee could not have drafted for. The document is not late because anyone was lazy. It is late because the drafting process is longer than the technology's release cycle, and that will remain true for every revision.
The second reason is that guidance is not implementation, and states are far better equipped to produce the first than the second. A framework on a state website changes nothing in a classroom until a teacher is trained to act on it. Here the data is sobering. RAND's American School District Panel found that only about 48 percent of districts had provided any AI training to teachers as of fall 2024, even as a majority of teachers were already using the tools. A state can publish guidance in a quarter. Building the professional development capacity to implement it across hundreds of districts takes years and money, and most guidance documents arrive with neither.
The third reason is local control. In most states, the state education agency genuinely cannot dictate classroom technology practice, and its guidance is advisory by design. That is why so many documents read as principles rather than procedures. The hard calls, which tools, which grade levels, which assignments, what disclosure looks like, are pushed down to districts, and districts vary enormously in their capacity to make them. The result is a patchwork where the quality of your AI guidance depends less on your state than on your zip code.
And that produces the fourth reason, the one I care about most: the lag is not evenly distributed. RAND found that 67 percent of low-poverty districts had trained teachers on AI by fall 2024, compared with 39 percent of high-poverty districts. The schools serving the students with the least access to these tools at home are also the schools least likely to have prepared their teachers. When guidance lags, it lags hardest exactly where the stakes are highest. A gap in policy becomes a gap in opportunity, and it is widening along the same lines as every other opportunity gap in American education.
The cost of waiting is already visible in the data
Some educators respond to the policy vacuum by holding still. No guidance, no action. I understand the instinct, particularly for teachers who have been burned by initiative whiplash before. But the data shows what holding still produces.
In RAND's 2025 surveys, more than 80 percent of students reported that their teachers had not explicitly taught them how to use AI for schoolwork. Read that carefully. It does not say students are not using AI. Student use is high and climbing. It says students are using it without instruction, which means they are learning their AI habits from each other, from social media, and from trial and error on graded assignments. The absence of guidance did not keep AI out of the classroom. It kept judgment out of the classroom.
The same pattern holds for teachers. The Gallup and Walton Family Foundation study, conducted in spring 2025 with more than 2,200 public school teachers, found that teachers who use AI weekly reclaim an average of 5.9 hours per week, yet 40 percent of teachers were not using AI at all. Across the nine categories of work tasks the survey examined, from lesson planning to drafting family communications, majorities of AI-using teachers, between 60 and 84 percent depending on the task, reported that the tools save them time. And the reclaimed hours were not vanishing into more paperwork: about 60 percent of teachers using these tools said they had more time to improve their teaching and give students individualized feedback.
Set those findings against the 40 percent of teachers using nothing, and the shape of the problem changes. This is no longer a story about early adopters and cautious colleagues making equally reasonable choices. The teachers who wait for official permission are working roughly six weeks a year longer than colleagues who did not wait, and the students of fluent teachers are getting explicit instruction in a skill that untrained classrooms leave to chance. Waiting is not neutral. It compounds, on both sides of the desk.
What teachers can do now
Everything that follows is legal, within a classroom teacher's existing authority, and consistent with every state guidance document I have read. None of it requires your state to move first.
1. Read the best available guidance, even if it is not your state's
If your state has guidance, read the actual document, not the press release, because the two often differ in usefulness. If your state has none, borrow. The strongest state frameworks are public, and they are anchored to the same recognized standards, most commonly the NIST AI Risk Management Framework and UNESCO's AI competency frameworks for students and teachers. An hour with one strong framework will give you a working vocabulary of the issues: data privacy, disclosure, appropriate use by task, human review of AI output. You do not need your own state's letterhead on the ideas to apply them responsibly.
When you read, read like a practitioner rather than a compliance officer. Skip past the preamble about transformation and go looking for three things. Does the document give concrete examples of acceptable and unacceptable use, or only principles? Does it name the standards and frameworks it is built on, so you can trace a recommendation back to its source? And does it say anything about professional learning, which is the tell for whether the authors expect the document to be implemented or merely cited? A document that fails all three tests can still teach you the vocabulary. A document that passes them is worth an afternoon and a highlighter.
2. Put your classroom policy in writing
The single highest-leverage move a teacher can make this summer is to write a one-page AI expectations statement for their own classroom. Define, in plain language, what AI use is encouraged, what is permitted with disclosure, and what is prohibited, and define it per assignment type rather than globally, because a blanket rule cannot survive contact with the difference between a brainstorm and a final essay. Share it with students on day one and with families in your first communication home. When district or state policy eventually arrives, you will adjust a document instead of inventing one, and in the meantime you have replaced ambiguity with a standard your students can actually follow.
A structure I have seen work in hundreds of classrooms is a simple three-tier disclosure model. Tier one: AI use is expected and part of the task, such as generating a draft to critique. Tier two: AI use is allowed with documentation, meaning the student notes what they asked and what they kept. Tier three: AI use is off limits because the point of the task is unassisted thinking, and you say so explicitly rather than assuming students will infer it. The tiers do the work that vague honor-code language cannot, because they replace a moral question, did you cheat, with an instructional one, did you follow the assignment's design. Students respond differently to the second question, and so do their families.
3. Teach AI use explicitly, because no one else is
That RAND finding, more than 80 percent of students never explicitly taught to use AI for schoolwork, is an instructional assignment addressed to us. Explicit instruction does not mean a semester course. It means showing students how you evaluate an AI response for accuracy, requiring them to document what they asked and what they changed, and building at least one assignment where critiquing AI output is the task itself. Students who can only accept an answer have learned extraction. Students who can interrogate one have learned judgment, and judgment is the literacy the next decade will pay for.
If you want a starting point that costs one class period, try this. Give students an AI-generated response to a question from your own curriculum, seeded with two or three errors you identified in advance. Their task is to find the errors, explain how they verified them, and rewrite the passage correctly with sources. In one period, students learn that fluent output is not the same as accurate output, that verification is a skill with steps, and that their teacher is neither afraid of the tool nor naive about it. That last lesson may be the most important one, because students take their cues about AI from how the adults around them handle it.
4. Claim the professional learning, and the hours, on your own schedule
The training gap is real, and it is not closing fast enough on its own. If your district is among the half that has not yet delivered meaningful AI professional development, you can still build the fluency yourself, and in most states documented professional learning counts toward the recertification hours you already owe. Choose learning that ends in classroom practice, not awareness. A certificate that changed nothing about your Monday is a receipt, not a credential. This conviction is the reason I built iTeachAI Academy the way I did, but the principle stands wherever you learn: the teachers who invest in their own AI fluency this year will be implementing whatever guidance finally arrives, while others are still decoding it.
5. Ask your district the questions state guidance should have answered
Even without a state framework, your district owes you clarity on a short list of questions, and asking them in writing tends to accelerate answers. Which AI tools are approved for use with students, and who vets them? What student data, if any, may be entered into an AI tool? What is our disclosure expectation for teacher-created and student-created work? Who do I ask when a new tool appears? If your district cannot answer these, you have just identified the agenda for its first AI committee meeting, which brings me to the last step.
6. Get a seat at the drafting table
There is also a funding argument to make while you are at it, and teachers are allowed to make it. Federal formula funds that districts already receive, Title II-A most directly, exist to improve teacher effectiveness, and high-quality professional development on instructional technology sits squarely within that purpose. When a district says there is no budget for AI training, what it usually means is that no one has written AI training into the plan for money that is already in the building. A teacher who asks the question in those terms, at a staff meeting or in a written request, changes the conversation from whether training is affordable to why it has not been scheduled.
Nearly every district will write or revise an AI policy in the next two years, and the difference between a usable policy and a press release is almost always whether practicing classroom teachers helped write it. The weakest guidance in this country was written about teachers. The strongest was written with them. Volunteer for the committee. If there is no committee, propose one to your principal with your one-page classroom policy attached as a starting draft. Policy is going to be written either way. The only question is whether the people who understand Monday morning are in the room.
The guidance will catch up. Your students cannot wait for it.
I want to end where the evidence points. State guidance is coming; the trajectory from zero states to two-thirds of them in roughly three years makes that clear, and the 2026 legislative session suggests the next wave will carry the force of law rather than advice. But every document that eventually arrives will describe a technology further along than the one it was drafted for, and every framework will still depend on a trained teacher to mean anything at all.
Which is to say the lag is permanent, and the answer to it has not changed since the first chatbot reached a school network: build capacity in the people closest to the students. Teachers who write their expectations down, teach AI use explicitly, document their own learning, and sit at the drafting table are not working around the system. They are how the system catches up. The state documents will keep arriving on committee time. Your classroom runs on Monday time, and Monday is the deadline that matters.
Janette Camacho, Ed.D., is the founder of iTeachAI Academy, a Google for Education Certified Trainer and Coach, FETC 2024/2025/2026 Featured Presenter, Adobe Creative Educator, Apple Teacher, and EdTech Digest 2026 Honoree. With 28+ years of K-12 classroom experience, she has facilitated AI professional development for educators across all 50 states.