Education AI Rules Are Already Law in Ten States

Europe's decision to push its high-risk education obligations to December 2027 is the subject of our main piece this week. Everywhere else, the direction of travel ran the other way: American statehouses have been setting district policy deadlines that have already come and gone, two new studies looked at how students learn to use AI rather than whether they cheat with it, and a district in rural New York stopped a robot purchase to write a data agreement first.

The Format of AI Training Changes What Survives Five Weeks Later

Pau Benazet i Montobbio, Janne Rotter and Davinia Hernández-Leo ran a quasi-experiment with 126 first-year engineering students, splitting them across two conditions for a single two-hour session on learning with generative AI. One group got the experiential version, hands-on work. The other got the classical lecture. Both were measured on metacognitive awareness before and after, then tracked across the trimester and assessed again five weeks later.

The immediate result favours the hands-on session. Experiential students came out ahead on engagement and on knowledge of cognition, the part that covers understanding which AI-use strategies actually work.

The five-week reading is the more useful one, and it points somewhere else. By then the two groups had converged on those same measures, so the immediate advantage had washed out entirely. What separated them instead was regulation of cognition, the harder skill of applying that knowledge in practice: the experiential group showed a delayed, continuous within-group increase that never appeared in the lecture group.

That has an awkward consequence for anyone evaluating a professional development session on using AI in education. Measure the week after and you capture a difference that does not last. Wait five weeks and the gain you find is a different one, in a different construct, moving in the opposite direction from the one you measured first. A two-hour lecture and a two-hour workshop cost the same, occupy the same slot on a compliance report, and do not produce the same student.

Source: arXiv

Identical AI Policies, Different Answers in Different Countries

Brian Harrington, Irina Zlotnikova, Gayathri Nadarajan and Samuel Ekundayo put a scenario-based survey to computing students at Canadian and South Korean universities, asking them to judge how ethical a set of AI-assisted coding practices were and whether each broke institutional rules. The institutions involved had functionally identical policies.

The answers were not identical. Canadian students were consistently more likely to read the same behaviour as both unethical and against policy than their Korean counterparts, with Mann-Whitney U tests and correlation coefficients showing significant differences across nearly all scenarios. The authors read the pattern through Hofstede's cultural dimensions, pointing at power distance, individualism and uncertainty avoidance as the factors shaping how students reason about permission in the first place.

One finding is directly usable by whoever has to write the policy. Across all the variables the researchers built into their scenarios, the one that moved ethical judgements most was the proportion of AI-generated code a student incorporated into the assignment. Degree, in other words, rather than category.

Most institutional guidance is built the other way round, as a list of permitted and prohibited activities. A policy that speaks in proportions gives students an anchor that survives the trip across a border; a policy that speaks in abstract integrity language gets re-interpreted by whoever is reading it. For any institution running multinational campuses or teaching large international cohorts, that gap is not a translation problem, and the authors argue the fix is guidelines written to be culturally responsive rather than merely uniform.

Source: arXiv

In the US, District AI Policy Deadlines Are Already Landing

FutureEd's legislative tracker counts 77 bills on AI in classroom instruction moving across 27 states this session, with ten states having enacted something. Several of those laws do the thing the EU has now deferred: they oblige individual institutions to have a policy, by a date.

Ohio got there first. Section 3301.24 of the Revised Code, enacted through the HB 96 budget bill, required every public school district, community school and STEM school in the state to adopt an artificial intelligence policy by 1 July 2026, a deadline that has now passed. The Department of Education and Workforce published its model policy in early January, covering acceptable use, privacy and FERPA obligations, vendor-evaluation standards, academic integrity, an AI workgroup requirement and periodic review. As the analysis from Kohrman Jackson Krantz stresses, the statute requires that a policy exist and says almost nothing about what it contains: a district can adopt the state model verbatim and be compliant, and nothing in the law requires it to teach AI or to use education AI tools at all. No enforcement mechanism is specified.

Maryland's Artificial Intelligence Ready Schools Act, SB 720, signed by Governor Wes Moore in May, works on a relative clock instead. The state department of education issues statewide guidance, and districts then have 120 days from its release to adopt aligned policies and designate AI coordinators, which puts the local deadline in the autumn. The guidance framework leans on data privacy, technology bias and keeping classroom instruction human-centred.

The rest of the set shows how differently states have answered the same question. Idaho's SB 1227 requires district policies aligned to a state framework spanning privacy, procurement safeguards, transparency, academic integrity, AI literacy standards and professional development. Oklahoma's SB 1734 gives districts until before the 2027-28 school year for written policies on approved uses, data protection, family transparency and periodic review. Virginia's HB 1186 requires alignment with state guidance. West Virginia's HB 5205 takes the most decisive line on non-compliance: state model policies apply automatically to any district that has not adopted a compliant local policy by July 2027.

That contrast is the part worth carrying. Ohio mandates a policy and names no consequence, which assumes districts act. West Virginia assumes some will not, and makes the state model the default. Both approaches produce a policy on paper; only one produces it reliably. And the obligations Ohio's model actually imposes, evaluating vendors and reviewing the policy on a schedule, are records questions before they are paperwork questions, since a periodic review is only as good as the record of what the system did in the interval, which is the kind of runtime audit trail AgentPMT keeps for the agents an institution runs itself.

An institution reading the European delay as breathing room should check its US map before relaxing. In six states, the deadline is either behind it or inside the next school year.

Source: FutureEd, Kohrman Jackson Krantz, Maryland Matters

Hiring Managers Say They'd Rather Buy AI Than Train a Graduate

Take the framing with the appropriate salt: this is a survey commissioned by ResumeTemplates.com, a commercial resume site, and it measures what 1,000 US hiring managers at companies with 101 or more employees say about their intentions rather than what any of them actually did. Higher Ed Dive's Lara Ewen reported it on 28 July.

The headline number is genuinely uncomfortable for anyone selling a credential: 48% of those managers said they would rather invest in AI tools than hire and train recent graduates. A majority have already moved some entry-level hiring budget towards AI, and close to a third said AI has reduced their need for graduate hires outright. Nearly half report asking one senior worker plus AI to cover work that used to take several entry-level people.

The same survey undercuts its own headline, though, and the correction deserves equal billing. Sixty-five percent still planned to hire the same number of 2026 graduates or more than the year before, and fewer than a quarter expected to hire fewer or none. Entry-level hiring is being reshaped rather than switched off.

What the managers complain about is the more actionable material, because almost none of it is about AI. Three quarters said recent graduates needed help understanding basic workplace documents, memos, contracts and budgets. Roughly two in five said they could not write a professional email, and a similar share found them short on data analysis and interpretation. A third named a lack of work ethic, only a small minority said they completely trusted a new graduate in front of a customer, and most would consider a graduate for certain roles only with internship experience behind them.

Read as a curriculum specification, that list is a gift. Document comprehension, professional written communication, and reading data well enough to argue from it are all teachable and all assessable, and an institution can start on them without waiting for a single policy question about AI to be settled.

Source: Higher Ed Dive

A New York District Paused a Robot Purchase to Write the Privacy Agreement First

Salamanca City Central School District sits in rural upstate New York, on the Seneca Nation of Indians reservation near the Pennsylvania border. It had committed a five-figure sum to Sally, a humanoid robot from the Las Vegas firm Realbotix, to support high school robotics and technology instruction: students would learn Arduino programming and robot maintenance, with a virtual AI teaching assistant and an at-home tutoring programme attached. Philip Marcelo reported for the Associated Press on 28 July that the plan is now paused.

Two distinct objections stopped it, and they are worth separating. State Education Commissioner Betty Rosa raised student data privacy in a letter to the district, and questioned the description of the robot as a tutoring platform given assurances that it would not deliver classroom instruction. Separately, New York State United Teachers president Melinda Person objected to the supplier's corporate family, saying a robot built by a company associated with sex dolls "has no business in our classrooms" and that students "need real relationships with caring adults." Realbotix responded that the educational unit was newly manufactured and purpose-built with no sex-doll components, and that its subsidiaries operate with separate management and operations.

Nothing was deployed here and nothing was measured, so this is not a story about a classroom robot that failed. It is a story about procurement diligence arriving before the purchase order clears rather than after, which is the sequence most districts manage in the other order.

The district has not cancelled. The pilot stays paused while it negotiates enhanced student data privacy agreements with state officials and runs community outreach, which is a description of the work that should precede any system touching student records. Superintendent Mark Beehler, who framed the purchase as extending technology access in an isolated community, also gave the clearest available answer to the question of whether AI will replace teachers: "There is no possible way a robot can replace a human in a school. Teaching is a human-to-human process."

Source: Associated Press via KSAT

Adoption Is Ahead of Guidance, and Educators Are Asking for the Guidance

Another vendor survey, and the vendor interest is direct: Instructure owns the Canvas LMS and sells into exactly the market it is describing. With that stated, the methodology is disclosed, which is more than most get. K-12 Dive's Anna Merod reported the results on 22 July, from 1,125 respondents polled online between 12 and 15 June, spanning K-12 and higher education educators, current and recent higher education students, and current K-12 parents.

Adoption is no longer the interesting variable. Around two thirds of educators use AI at least occasionally and nearly three quarters of parents say their children use it. The appetite for prohibition is much smaller than the volume of the debate suggests: only about a third of educators favour restricting AI use, and parents are no keener.

Parents are concerned, but specifically rather than generally. Roughly six in ten worry about academic integrity and plagiarism, about as many about a loss of critical thinking, and nearly as many about overreliance on technology. Those are three different problems with three different answers, and only the first one is the argument most institutional policy is written to win.

The gap is training. Forty-five percent of educators have had none at all, most of the rest describe theirs as partial, and fewer than one in ten call it comprehensive. Instructure's own reading is that people are not waiting for permission to use AI, they are looking for practical guidance and clearer boundaries, and its recommendations run to AI literacy instruction, ongoing professional development, explicit boundaries on education AI tools, and building evidence of impact over time.

Which loops back to where this digest started. Districts in Ohio have a policy deadline behind them and districts in Maryland have one coming in the autumn, and both arrive at the same next problem: the nearly half of educators who have had no AI training at all, and the majority whose training stopped short of comprehensive. The engineering-student study says the format of that training decides what is still there five weeks later. Institutions working through AI in education have the unusual luxury this month of knowing both what to build and roughly how to teach it.

Source: K-12 Dive

Sources

• Experiential Versus Instructional Approaches for Eliciting Metacognitive Awareness in AI-Assisted Learning, arXiv
• Did Alice Do Wrong? Cross-Cultural Differences in Student Perceptions of Generative AI Use in University Computing Education, arXiv
• Legislative Tracker: 2026 State AI in Education Bills, FutureEd
• Ohio's July 1, 2026 School AI Policy Deadline, Kohrman Jackson Krantz
• Maryland school districts face fall deadline to set AI policies, Maryland Matters
• New grads have to compete with AI for entry-level roles, hiring managers say, Higher Ed Dive
• New York school pauses plan to deploy humanlike AI robot teacher after backlash, Associated Press via KSAT
• AI embraced by more students and educators, Instructure finds, K-12 Dive

Read the full report: Europe Delays Education AI Grading Rules to 2027