.jpg)
The States Are Not Waiting
While the federal government continues to debate how, or whether, to regulate artificial intelligence, the most consequential decisions about AI in education are being made in state capitals. In the span of a few weeks this spring, Utah committed to putting Google Gemini in front of every public school student, Maryland enacted a law forcing districts to write AI policies, and the State University of New York mandated AI literacy across 64 campuses. None of these waited for Washington. The states have decided that the technology is already in classrooms and that governing it cannot wait for a federal framework that may never arrive.
This bottom up dynamic is how American education policy often works, but the stakes here are unusually high. The choices states make now, which tools to deploy, which guardrails to require, which skills to mandate, will shape how an entire generation encounters AI. We are watching the early formation of a patchwork, and the contours of that patchwork will determine whether AI in schools becomes a thoughtful augmentation of learning or an unmanaged experiment run on children at scale.
Utah Goes All In
Utah has made the boldest deployment bet. The state announced a statewide Google Gemini rollout across all public schools beginning in the 2026-27 school year, reaching roughly 680,000 students and 28,000 educators at no cost. The scale is the headline: this is not a pilot in a handful of districts but a commitment to make a specific commercial AI assistant part of the standard toolkit for an entire state's public education system. It is one of the largest single deployments of a generative AI tool in K-12 anywhere in the country.
The economics deserve scrutiny precisely because the price is zero. A vendor providing AI to 680,000 students at no cost is making an investment, and the return is not philanthropy. It is mindshare, data, and the habit formation that comes from being the assistant a student learns on. We do not say that to impugn Utah's choice, which addresses a real demand, but to flag what districts everywhere should ask: when a frontier AI tool is free at the point of use, what is the vendor actually getting, and what governance surrounds the student data that flows through it.
Maryland Writes the Rules
Where Utah leads with deployment, Maryland leads with governance. The state's AI Ready Schools Act requires districts to adopt AI policies within 120 days and to designate AI coordinators by June 1, 2027. The law does not pick a tool or a vendor; it forces every district to confront the questions that deployment alone skips, acceptable use, data protection, academic integrity, and accountability, and to assign a human owner responsible for answering them. It is a policy first approach to a technology that has mostly arrived policy last.
This is the more durable model, in our view, because it builds the muscle rather than buying the tool. Technology will change; the Gemini of 2026 will not be the assistant of 2030. A district that has a coordinator, a policy, and a process for evaluating AI is equipped to adapt as the tools evolve. A district that simply deployed whatever was free and free is not. Maryland is betting that the scarce resource in education AI is not access to models but institutional capacity to govern them, and that bet looks right.
Higher Education Joins In
The shift is not limited to K-12. The State University of New York adopted a formal systemwide AI policy across all 64 campuses, mandating AI literacy in general education for all incoming undergraduates starting in Fall 2026. That is a significant statement from one of the largest public university systems in the country: AI fluency is now a baseline graduation competency, not an elective curiosity. SUNY is treating the ability to use AI critically and responsibly as a core part of what a degree should certify.
Pairing literacy requirements with deployment is the combination that matters. Putting tools in students' hands without teaching them to interrogate AI output produces dependence, not capability. SUNY's approach acknowledges that the goal is not students who can prompt a chatbot but students who understand what the chatbot is doing, where it fails, and when to distrust it. If the K-12 deployments in states like Utah are eventually matched by literacy mandates like SUNY's, the pipeline could produce graduates genuinely prepared for an AI saturated workforce rather than merely habituated to one assistant.
A Cautionary Tale
The enthusiasm should be tempered by evidence, and a story out of North Carolina supplies the caution. State legislators defended a 10 million dollar earmark for Khan Academy's Khanmigo AI tutor even as founder Sal Khan candidly acknowledged that the tool was a non event for many students who simply did not use it much. Democratic senators questioned the evidence base for the spending, and the questions were fair. Buying an AI tutor is easy; getting students to use it well, and proving that the use improves learning, is hard and largely unproven at scale.
This is the gap between procurement and outcomes that every education AI decision has to bridge. The most well intentioned deployment fails if the tool sits unused or if its benefits cannot be demonstrated. We would urge every state writing a check or signing a deal to fund evaluation alongside deployment, because the alternative is spending public money on the assumption that access equals impact. Khanmigo's candid admission, from one of the more credible names in the space, is a useful reminder that adoption is not the same as effectiveness.
The Patchwork Ahead
The result of all this state level activity is a fragmented landscape, and fragmentation has costs. A vendor selling into Utah faces different requirements than one selling into Maryland, and a family moving between states will find their children governed by different rules entirely. For edtech companies, the compliance burden of a fifty state patchwork is real and rising. For districts, the absence of a common standard means reinventing governance locally, often without the expertise to do it well. The lack of a federal floor is not a neutral fact; it pushes complexity and cost downward.
Yet the patchwork also has a virtue: it is a national experiment running in parallel. Utah's all in deployment, Maryland's policy first mandate, and SUNY's literacy requirement are testable hypotheses about how to bring AI into education responsibly. Within a few years we will have evidence about which approach produced better outcomes and fewer harms. The states moving first are taking on risk on behalf of students, and the least the rest of the system can do is learn from what they find. The terms of classroom AI are being set right now, and they are being set in the states.
