A few years from now, a child asks a question to the children's AI handed out at school. What comes back is a short, harmless answer and one line: "Why not ask your teacher?" After a few more tries, the child may mutter, "AI isn't very smart, is it?"
But it isn't AI that isn't smart. It is the watered-down AI made for children that the child touched. And today, most of the real AI tools exclude elementary school children in their terms of use.
Japan will teach "the ability to make full use of AI" from elementary school
On October 5, 2026, the Curriculum Subcommittee of Japan's Central Council for Education published its summary of deliberations toward the next Courses of Study. It proposes adding an "information domain" to the integrated inquiry time in elementary schools and creating a new junior high school subject, "Information and Technology." Full implementation is expected in fiscal 2030 for elementary schools and fiscal 2031 for junior high schools, with phased early implementation from fiscal 2028 also under consideration.
The summary places "the ability to make full use of AI" at the core of information literacy. It defines this not as "mere operating skills" but as the ability to use AI appropriately to "discover and solve problems and form one's own ideas." It also acknowledges that children already come into contact with AI every day "through search engines and the like."
Yet the leading AI tools shut out elementary school children
Anthropic's Claude requires users to be 18 or older under its terms of use. OpenAI's ChatGPT and Codex require users to be 13 or older, and those under 18 need a parent's permission. Google's Gemini can be used by children under 13 if a parent allows it through Family Link, but such routes are still the exception. There is almost no cutting-edge AI that elementary school children can use with their own hands.
Japan's Ministry of Education guidelines on generative AI also ask schools to make students follow each service's terms, including age limits. They say direct use by elementary school children "requires more careful judgment," and give as an example showing children a teacher's conversation with AI. A direction for revising the guidelines, presented in October 2026, includes a proposal to add the need for "AI specialized for education."
The government says it will build the ability to use AI, while the tools themselves leave children out. Children are the ones to be taught, yet they are kept away from the real thing. That is a serious gap.
The same thing happened with computers
Something similar has already happened once. Among the learning devices distributed under the GIGA School program, ChromeOS accounts for 60% in the second phase now being rolled out, while Windows has only 10% (MM Research Institute, July 2025). Meanwhile, of web use from computers in Japan, Windows accounts for about 80% and ChromeOS for less than 1% (StatCounter, September 2026).
The tools children get used to at school differ from the tools they will face in society. Given cost and ease of management, schools had their reasons. Even so, whether a tool is easy for adults to manage has often come before whether children will use it as adults. We must not repeat the same mistake with AI.
There are real reasons for drawing the line
The age limits have reasons behind them. The line at 13 reflects a U.S. law (COPPA) that strictly governs the handling of personal information of children under 13. In October 2025, the U.S. company Character.AI, which runs a conversational AI app, announced, in response to news reports and questions from regulators, that it would end open-ended chats with AI for users under 18 by November 25. Children do need protection.
But shutting children out by age alone removes the chance to learn along with the danger. Protecting children and keeping them from using AI are not the same thing.
Using the real thing is the shortest path
Watching a teacher talk with AI on a screen is a completely different experience from asking your own questions and checking the answers that come back. That AI can be wrong in convincing ways, and that answers change when you ask differently, only sink in when you use it yourself. If children know only AI with its functions cut back, they will grow up believing "this is all AI can do." The AI children will use as adults is whatever is newest at that time.
What we need is not a system that closes the door by age, but one that opens up the latest AI on the condition that an adult is nearby. We hope AI companies will offer ways for children to use their latest models under the watch of parents and teachers. Schools and communities need places where children can touch the real thing with an adult beside them.
There is something families can do starting today. A parent operates the AI on their own account and looks at the screen together with the child. The parent types in the child's question, and after reading the answer, asks, "Which part did you understand?" and "How could we check this?" Instead of using the answer as it is, explain it again in your own words. This also matches the summary's warning against using AI output as it is, "without putting it into words."
Digital Kodomo BASE runs the free After-School Lab in Ota City, Tokyo, as a place where children can touch real tools (we are currently accepting visits and waitlist applications). We also continue to deliver computers free of charge to children who don't have one at home.