The number that should worry every procurement officer
The K-12 generative AI market grew from essentially nothing to about $730 million in 2026, according to industry data reported this month, with projections putting it at $18.5 billion by 2036 on roughly 40 percent annual growth. That trajectory is faster than almost any prior K-12 technology category, including the one-to-one device rollouts and the first wave of learning management systems. Adoption is already near saturation on the user side: 85 percent of teachers and 86 percent of students report using generative AI tools in schoolwork.
A market moving that fast, with usage that far ahead of adoption, is exactly the environment where governance structures fall behind. School districts are used to procurement cycles measured in years, built around RFPs, pilot programs, and board approval. A market growing 40 percent annually does not wait for that cadence, and vendors know it. The result is a large and growing share of classroom AI use happening through free tiers and informal adoption rather than through anything a district technology office formally vetted.
Free access today, lock-in tomorrow
Part of what is driving usage ahead of governance is simple pricing strategy. OpenAI has made ChatGPT free for schools through June 2028, and Anthropic has done the same with Claude through June 2027. Google has folded Gemini and NotebookLM directly into Google Classroom, a platform many districts already run their entire instructional workflow through. Smaller players like MagicSchool, SchoolAI, and Bloomy are competing for the same teachers and students with narrower, purpose-built tools.
Multi-year free access windows are a deliberate strategy to embed a tool into daily instructional habits before the district ever negotiates a paid contract or runs a competitive procurement process. By the time a free tier expires, a school's grading workflows, lesson planning habits, and student familiarity may already depend on that specific product, which weakens the district's negotiating position and makes a genuine build-versus-buy evaluation much harder to run honestly.
What the research actually shows
The evidence base has not kept pace with the spending or the adoption curve. Stanford researchers have documented what they call a performance cliff: students using AI tools without restriction show immediate gains on assigned work, then perform worse than baseline once the tool is taken away, suggesting the tool is substituting for learning rather than accelerating it. Chris Agnew, who leads Stanford's Generative AI for Education Hub, says the effectiveness of personalized-learning chatbots specifically remains unproven by rigorous, randomized controlled trials, despite that being the central pitch most vendors lead with.
Safety evaluation has also lagged behind deployment. Common Sense Media rated Google's Gemini for schools as high risk, finding that it applies the same content and interaction rules to a sixth grader as it does to a high school senior, and that it can surface content inappropriate for younger students. Separate research on AI-assisted grading has found inconsistent scoring and bias correlated with names associated with different races and genders, a finding with direct legal exposure for any district using AI-scored assessments in ways that affect grades or placement decisions.
When procurement goes wrong
The governance gap has already produced consequences serious enough to end careers. Los Angeles Unified Superintendent Alberto Carvalho resigned in June 2026 following an FBI raid connected to an investigation into an AI chatbot contract, one of the most senior departures directly tied to AI procurement in K-12 history. New York City's schools chancellor separately paused all AI software purchases after the City Council demanded a moratorium pending review, a sign that even the country's largest district lost confidence in its own vetting process mid-cycle.
Neither incident is really about a single bad vendor. Both point to the same structural problem: districts are signing AI contracts faster than their procurement and legal review processes were designed to handle, using evaluation criteria built for traditional software rather than tools whose behavior changes with every model update. A contract that looked defensible at signing can become a liability months later if the underlying model's behavior shifts and nobody in the district is positioned to notice or renegotiate.
The states writing the rules Washington won't
Federal policy has stayed encouragement rather than enforcement. The current administration has pushed districts toward AI adoption without attaching binding vetting standards, and more than 30 states have adopted some form of AI guidance for schools, though the rigor varies enormously. Wyoming and North Carolina stand out for building genuinely detailed frameworks. Wyoming's education department produced a public 'threat assessment' video specifically flagging red flags for districts to watch for in vendor tools: excessive data collection, embedded advertising targeting, and evidence of learning harm.
That two mid-sized states have done more substantive work than the federal government or most larger states says something about where real vetting capacity currently sits, which is unevenly and mostly by accident of which state education agency happened to prioritize it. A district in a state without a rigorous framework is, in practice, on its own for evaluating vendor claims, data practices, and safety design, with no state-level backstop to catch a bad contract before it is signed.
What buyers should demand before they sign
District technology and procurement leaders should treat every AI vendor pitch with the same skepticism they would apply to any enterprise software claiming measurable outcomes without independent verification. That means asking specifically for randomized controlled trial data, not case studies, before accepting a personalization or tutoring claim, and asking for a third-party safety and content-risk assessment, not just a vendor's self-reported compliance statement, before deploying any tool with direct student interaction.
It also means building contract terms that survive a model update: the right to re-evaluate and exit if a vendor changes the underlying model powering a tool, clear data retention and deletion terms independent of the free-tier period, and a designated internal owner responsible for AI vendor review who is not the same person managing day-to-day procurement volume. The market is not going to slow down to match district review cycles. The only lever districts actually control is how rigorously they use the cycles they have.
