Over a single late-July weekend in Boston, 98 teenagers from all 50 states sat down in a precise replica of the U.S. Senate chamber and did something the real Senate has not managed: they wrote rules for artificial intelligence in schools, argued them out, and passed them. Their bill, the Students First Act, cleared the floor 82 to 16. It is not law, and the teens are not senators. But the exercise, hosted by the Edward M. Kennedy Institute, the School Superintendents Association, and the MIT-linked nonprofit Day of AI, produced something with a strange amount of real-world weight.
The reason is the vacuum it was written into. There is no national policy on how American schools should handle AI, so tens of thousands of districts have been improvising their own. The superintendents' association now plans to send the students' text to its 10,000 members, many of whom are drafting AI rules right now with little to guide them. A model bill written by teenagers may end up shaping how actual classrooms operate, precisely because the adults never produced one.
What the students wrote is more careful than the stunt might suggest. The bill bans AI on all graded tests, forbids using it to do the actual writing, but allows it for editing, brainstorming and studying after eighth grade. Students who use it must cite how, and prove their grasp through discussion, handwritten work or an oral defence. Crucially, they refused to trust the software: any student flagged by an AI detector must be reviewed by two human officials, because, as the teens noted, the detectors are unreliable. Teachers, in turn, must post a clear AI policy each semester and personally investigate before reporting anyone.
The debates cut deeper than plagiarism. One of the sharpest exchanges was about a student turning to a school chatbot for emotional support. Several worried that a teenager confiding about their sexuality or a mental-health crisis deserved privacy. But Dylan Dornack of Iowa argued that a student in genuine danger needs a human alerted, not a bot keeping the secret. The chamber sided with him: when a chatbot detects a student in crisis, it must not simply answer, it must tell a trusted adult. That is a more thoughtful position than many commercial products have reached.
It was not unanimous or naive. West Virginia's Ethan Liu nearly torpedoed the bill over its contradictions, complaining it both restricted students and handed teachers vague discretion, and he had a speech ready against it. Organisers let the students patch the worst inconsistencies at the last minute, and Liu changed his vote. Alabama's Jadelynn Petitjean still called the result an "unstable foundation" and voted no. That friction is the point; a bill nobody argued over would be worth less.
The animating fear was memorably plain. "We missed the mark with social media and phones," said Ohio senior Hunter Wurzel, describing a generation that watched adults fail to manage the last technology to reshape their childhoods. This time, the students decided, they would rather write the rules than inherit them. It is a gentle embarrassment that the most coherent AI-in-schools policy of the summer came from people too young to vote on the real thing. It is also, quietly, a good sign.