Researchers and commentators have described this pattern as evidence of "cognitive offloading" — students use AI to complete assignments efficiently without building the underlying knowledge needed for closed-book assessments . The losses were largest among high-achieving students, suggesting that even strong students are vulnerable to the learning penalty when AI is used as a shortcut .

Homework Shortcut vs. In-Class Tutoring Tool

The study's core implication is that AI's impact depends entirely on the context of use. When students rely on generative AI to finish homework faster and get better homework grades, they bypass the effortful practice and productive struggle that builds long-term retention. Exams, which measure durable knowledge, reveal the gap.

This finding aligns with a separate large-scale analysis of 3.2 million math learning interactions on the ALEKS platform. That study found that after ChatGPT's release, learning time on AI-susceptible problems declined 2.8% per quarter among college students, cumulating to 26.9% over eleven quarters. High schoolers showed a 31.3% decline, middle schoolers 9.0%, while Grade 5 students showed no detectable change .

The critical distinction is clear: AI used as a replacement for student thinking degrades learning, whereas AI used as a supplement under teacher guidance may support inquiry-based learning without the same risks. A 2025 meta-analysis of 19 studies found that students with teacher support in student-GenAI interactions had significantly larger academic gains (g = 1.426) than those without teacher support (g = 0.078) .

Global Policy Responses: China

China adopted a tiered, regulated approach rather than an outright ban. In May 2025, the Ministry of Education issued two guidelines promoting the scientific and regulated use of AI in kindergartens and primary and secondary schools . The guidelines established clear age-based rules:

• Primary school students are prohibited from independently using open-ended generative AI tools for completing assignments .
• Middle school students may explore AI-generated content under supervision .
• High school students can engage in inquiry-based learning that incorporates AI's technical principles .

Teachers are also restricted: they may not use generative AI as a substitute for their core teaching responsibilities or as a replacement for human-led teaching . The guidelines explicitly ban students from directly copying AI-generated content as answers to homework or examinations .

This reflects a strategy of integration with guardrails. China simultaneously pushed to embed AI literacy into its basic education curriculum, with a March 2025 plan requiring at least eight class hours annually per student from primary through high school .

Global Policy Responses: Norway

Norway has taken the most restrictive stance among Western nations. On June 19, 2026, Prime Minister Jonas Gahr Støre announced a near-total ban on generative AI for students in grades 1–7 (ages 6–13), taking effect at the start of the August 2026 school year .

The policy details:

• Grades 1–7 (ages 6–13): Generative AI is effectively banned for educational assignments .
• Grades 8–10 (ages 14–16): Limited, strictly supervised use is permitted under teacher guidance .
• Grades 11–13 (ages 17–19): Emphasis is on learning to use AI appropriately for future employment and higher education .

Støre stated at a press conference that using AI increases the risk that young children "skip important steps in their education" . Officials explicitly connected the policy to a broad decline in educational test scores, which had previously prompted a 2024 smartphone ban in Norwegian schools .

Global Policy Responses: United States

The United States has no federal K-12 AI policy, creating a fragmented state-by-state landscape. As of 2026, 134 AI-in-education bills have been introduced across 31 states . Key developments include:

• New York proposed a ban on AI use in classrooms below ninth grade (Assembly Bill A.9190) .
• Ohio gave every K-12 district until July 1, 2026 to adopt a formal AI policy or face being out of compliance with state law .
• Tennessee now requires every school district to adopt a formal AI policy .
• Oklahoma mandates educator supervision and parent disclosure for AI use (SB 1734) .
• South Carolina requires written parental opt-in before any student interacts with an AI tool, and prohibits AI from replacing licensed teachers .

At the federal level, a December 2025 White House executive order sought to eliminate "state law obstruction" of national AI policy, creating tension with state-level restrictions and leading to the establishment of an AI Litigation Task Force within the Department of Justice .

The Emerging Consensus

The converging evidence from the Chinese study, the ALEKS math data, Norwegian policy, and U.S. state legislation suggests an emerging global consensus: unsupervised generative AI use by younger students undermines learning by short-circuiting the effortful practice that builds durable knowledge.

The policy debate is shifting from "should we allow AI in schools?" to "under what conditions and at what ages?" China, Norway, and various U.S. states have arrived at different answers along the same spectrum of tiered, age-dependent restriction — but they share a common conclusion: unsupervised AI use is particularly harmful for younger students, and teacher oversight is essential for any beneficial application.

Researchers and commentators have described this pattern as evidence of "cognitive offloading" — students use AI to complete assignments efficiently without building the underlying knowledge needed for closed-book assessments . The losses were largest among high-achieving students, suggesting that even strong students are vulnerable to the learning penalty when AI is used as a shortcut .

Homework Shortcut vs. In-Class Tutoring Tool

The study's core implication is that AI's impact depends entirely on the context of use. When students rely on generative AI to finish homework faster and get better homework grades, they bypass the effortful practice and productive struggle that builds long-term retention. Exams, which measure durable knowledge, reveal the gap.

This finding aligns with a separate large-scale analysis of 3.2 million math learning interactions on the ALEKS platform. That study found that after ChatGPT's release, learning time on AI-susceptible problems declined 2.8% per quarter among college students, cumulating to 26.9% over eleven quarters. High schoolers showed a 31.3% decline, middle schoolers 9.0%, while Grade 5 students showed no detectable change .

The critical distinction is clear: AI used as a replacement for student thinking degrades learning, whereas AI used as a supplement under teacher guidance may support inquiry-based learning without the same risks. A 2025 meta-analysis of 19 studies found that students with teacher support in student-GenAI interactions had significantly larger academic gains (g = 1.426) than those without teacher support (g = 0.078) .

Global Policy Responses: China

China adopted a tiered, regulated approach rather than an outright ban. In May 2025, the Ministry of Education issued two guidelines promoting the scientific and regulated use of AI in kindergartens and primary and secondary schools . The guidelines established clear age-based rules:

• Primary school students are prohibited from independently using open-ended generative AI tools for completing assignments .
• Middle school students may explore AI-generated content under supervision .
• High school students can engage in inquiry-based learning that incorporates AI's technical principles .

Teachers are also restricted: they may not use generative AI as a substitute for their core teaching responsibilities or as a replacement for human-led teaching . The guidelines explicitly ban students from directly copying AI-generated content as answers to homework or examinations .

This reflects a strategy of integration with guardrails. China simultaneously pushed to embed AI literacy into its basic education curriculum, with a March 2025 plan requiring at least eight class hours annually per student from primary through high school .

Global Policy Responses: Norway

Norway has taken the most restrictive stance among Western nations. On June 19, 2026, Prime Minister Jonas Gahr Støre announced a near-total ban on generative AI for students in grades 1–7 (ages 6–13), taking effect at the start of the August 2026 school year .

The policy details:

• Grades 1–7 (ages 6–13): Generative AI is effectively banned for educational assignments .
• Grades 8–10 (ages 14–16): Limited, strictly supervised use is permitted under teacher guidance .
• Grades 11–13 (ages 17–19): Emphasis is on learning to use AI appropriately for future employment and higher education .

Støre stated at a press conference that using AI increases the risk that young children "skip important steps in their education" . Officials explicitly connected the policy to a broad decline in educational test scores, which had previously prompted a 2024 smartphone ban in Norwegian schools .

Global Policy Responses: United States

The United States has no federal K-12 AI policy, creating a fragmented state-by-state landscape. As of 2026, 134 AI-in-education bills have been introduced across 31 states . Key developments include:

• New York proposed a ban on AI use in classrooms below ninth grade (Assembly Bill A.9190) .
• Ohio gave every K-12 district until July 1, 2026 to adopt a formal AI policy or face being out of compliance with state law .
• Tennessee now requires every school district to adopt a formal AI policy .
• Oklahoma mandates educator supervision and parent disclosure for AI use (SB 1734) .
• South Carolina requires written parental opt-in before any student interacts with an AI tool, and prohibits AI from replacing licensed teachers .

At the federal level, a December 2025 White House executive order sought to eliminate "state law obstruction" of national AI policy, creating tension with state-level restrictions and leading to the establishment of an AI Litigation Task Force within the Department of Justice .

The Emerging Consensus

The converging evidence from the Chinese study, the ALEKS math data, Norwegian policy, and U.S. state legislation suggests an emerging global consensus: unsupervised generative AI use by younger students undermines learning by short-circuiting the effortful practice that builds durable knowledge.

The policy debate is shifting from "should we allow AI in schools?" to "under what conditions and at what ages?" China, Norway, and various U.S. states have arrived at different answers along the same spectrum of tiered, age-dependent restriction — but they share a common conclusion: unsupervised AI use is particularly harmful for younger students, and teacher oversight is essential for any beneficial application.