New York City Public Schools this week imposed a one-year moratorium on student-facing generative artificial intelligence for students in prekindergarten through eighth grade, a policy that is being paired with overall screen time restrictions and will affect nearly 600,000 students. A limited number of the city's high school students willhave some controlled access, including required AI literacy lessons and selected, monitored AI pilots, though
Los Angeles Unified School District, meanwhile, has issued its own ban, restricting students from accessing generative AI tools on district-issued devices for the 2026-27 school year while the nation's second-largest school district develops a broader approach to the technology. Teachers will continue to have access.
And Florida is currently debating a plan that all of its public schools — across K-12 and colleges and universities — to adopt policies governing the use and limits of artificial intelligence.
The timing is fascinating. Schools and colleges at every level have spent the past several years scrambling to incorporate generative AI into teaching, learning and administrative work, often before developing consistent policies governing when students should — or should not — use it. Previous reporting by The EDU Ledger has documented that uneven policy landscape across higher education.
But many K-12 districts are beginning to grapple with the same question some institutions are asking: If students are supposed to be learning how to think, write, reason and solve problems, when does using AI become a substitute for learning rather than a tool for it?
The New York City announcement acknowledged outright that concerns about student learning were a major reason for the ban. Still, the district is not rejecting AI outright. The high school pilot will provide instruction on how AI works, data privacy, bias and appropriate use. Career-readiness programs can also incorporate approved AI tools when doing so is necessary to develop workforce skills.
The policy seems to recognize that while generative AI cripples learning, students still need to understand AI without necessarily being permitted to use it to complete the work through which they are supposed to acquire other skills.
Higher education is confronting the same problem, although generally with more granular rules. A 2025 study in the International Journal of Educational Technology in Higher Education examined generative-AI guidelines at the top 50 U.S. universities and found that most institutions have avoided a single, campus-wide rule in favor of course- and instructor-level discretion. The researchers reported that 94% of the universities studied had faculty guidance emphasizing the need to establish and communicate clear expectations — but left the actual determination of whether AI is permitted, restricted, or prohibited for a given assignment up to individual instructors.
Columbia Law School's new 2026-27 AI policy does not impose a blanket ban but says students must retain intellectual responsibility for work submitted for academic credit. AI may be used as a learning aid or to test ideas, but it cannot replace the student's legal reasoning or serve as an undisclosed ghostwriter. The policy prohibits AI from performing the legal analysis or judgment that an assignment is intended to assess and generally prohibits its use during examinations.
At the University of California, Berkeley School of Law, however, students are prohibited from using AI to conceptualize, outline, draft, revise, translate, or edit work submitted for credit.
Traditionally, academic integrity policies have focused on whether students received unauthorized assistance or represented someone else's work as their own. Generative AI complicates that framework because the assistance can be instantaneous, personalized, and difficult to distinguish from a student's own work. And increasingly, institutions cannot reliably solve that problem by simply trying to catch AI-generated work after the fact.
Recent research has raised continued concerns about the reliability of AI detectors in high-stakes academic settings. A June study comparing four widely used AI-detection tools found substantial differences in their ability to identify AI-generated and hybrid human-AI writing and concluded that detection tools should not be used as the sole evidence in high-stakes decisions. And an August paper published in Science Direct found that 13 detection systems showed systematic failures across student assignments, theses and code, concluding that detector performance remained inadequate for high-stakes assessment.
Not only that, but a 2023 study by Stanford researchers found that seven widely used AI detectors incorrectly classified human-written essays by non-native English speakers as AI-generated at an average rate of 61.3%, raising concerns that relying on automated detection in academic-integrity cases could disproportionately subject multilingual students to false accusations of misconduct.
For many institutions, the answer is, instead of trying to determine whether AI did the work, faculty members are working to design learning experiences in which students are required to do the work themselves.
The implications extend well beyond writing assignments.
A nursing program may need students to learn clinical reasoning before allowing an AI system to help them organize information. A teacher-preparation program may need candidates to demonstrate that they can develop a lesson before using AI to generate alternatives. A computer science program may reasonably teach students how to work alongside coding assistants while still requiring them to demonstrate that they understand the underlying code. A law student may use AI to test an argument while remaining responsible for constructing and defending it.
The challenge for educators, then, is not simply deciding whether AI is good or bad, but what the student is supposed to learn from the assignment — and whether AI helps the student learn it or does it for them.
That distinction may also help explain why the emerging policy landscape looks less like a wholesale rejection of AI than a search for boundaries. A recent analysis of university policies found that institutions are largely choosing adaptable guidance over permanent, institution-wide rules. The common thread is an effort to balance AI literacy with assessment integrity, instructor autonomy and preparation for an increasingly AI-dependent workplace.
While many school districts’ concerns — spurned by parent advocacy — center around developmental appropriateness of the tools on young minds that are still being formed, college and university students are closer to — and often already in — the workforce, where AI proficiency may be expected. At the same time, colleges are responsible for certifying that graduates actually possess the knowledge and abilities their degrees represent.
As institutions walk back their initial rush to adopt generative AI toward more deliberate policies, that may become the central academic-integrity question of the AI era: not whether a student touched the technology, but whether the student's work demonstrates what the institution intended the student to learn.
