Should Schools Teach Students How AI Works, Not Just How to Use It?

The partnership contains five concrete programmes

Over the following year, OpenAI and CodeAI say they will establish a joint advisory council involving specialists in child development, youth public policy, and learning science. The council is meant to provide ongoing advice about emerging risks, educational needs, responsible use, and the development of ChatGPT for Teens. Its value will depend on who holds decision-making power and whether recommendations and responses become visible to the public.

The partnership will also use CodeAI's Hour of AI to introduce millions of students to foundational AI literacy; run a first Builders Challenge in which high-school students create projects, receive mentorship from OpenAI staff, and present work alongside teachers; and provide OpenAI expertise to the free, year-long AI Foundations high-school course developed by CodeAI.

A fifth programme, Career Journeys, will connect students with OpenAI researchers, engineers, and leaders to show how AI-related work is done. Together, these initiatives cover exposure, building, curriculum, professional examples, and governance. They are more substantial than a one-day prompt workshop, though the announcement does not yet report participation by geography, completion standards, assessment methods, or outcomes.

Only 16% of high-school leaders say all their students learn the technical knowledge needed to understand AI, while 75% of students say that understanding will matter more in their future.

AI literacy must go beyond prompt technique

A student who can obtain a polished answer is not necessarily AI literate. Useful literacy begins with a working mental model: language models predict patterns rather than retrieve guaranteed truth; outputs can vary with wording and context; training data can carry social bias; and a fluent explanation can include fabricated details or citations.

Students should practise verification as part of using the tool. That includes tracing a claim to an inspectable source, comparing output with primary material, identifying uncertainty, and recognising tasks where an AI system lacks sufficient context. They should also understand the data question: what information should never be entered, what may be retained, and how school or provider policies affect privacy.

The goal is not to turn every student into a machine-learning engineer. It is to give young people enough technical and social understanding to question a system's authority. Karim Meghji, CodeAI's CEO, frames that outcome directly in the announcement: students should be able to catch mistakes and know when to stop trusting the technology.

Teachers need time, examples, and authority—not just access

OpenAI's announcement says teachers should play a meaningful role in how AI enters classrooms and connects the new partnership to ChatGPT for Teachers and work with the American Federation of Teachers. That principle is important because educators are responsible for assessment, safeguarding, subject accuracy, and the relationships in which learning occurs.

Professional development must therefore cover more than product navigation. Teachers need examples of failure in their subject area, guidance for redesigning assignments, clear escalation routes for harmful output, and enough planning time to decide when AI use supports a learning objective. Otherwise, schools risk adding tools while transferring extra monitoring labour to already stretched staff.

Assessment requires particular attention. If a task can be completed by pasting a question into a chatbot, schools can respond with surveillance or redesign the task around process: source logs, oral explanation, drafts, reflection, and in-class application. AI literacy becomes credible when students must show how they reached a conclusion, not only submit the conclusion.

A US partnership still leaves a global access question

The announcement speaks broadly about the first generation growing up with AI, but its examples—a national student challenge, US high-school leadership data, and existing domestic education partnerships—are rooted primarily in the United States. Students elsewhere face different curricula, languages, data costs, connectivity, teacher capacity, and rules about technology providers.

A programme intended to reach millions should publish where it reaches, which students complete it, how learning is measured, and whether materials work on low-bandwidth devices. It should also make room for independent educators and civil-society organisations to evaluate safety claims rather than allowing a product company to define responsible use alone.

The strongest part of the partnership is its stated ambition to teach judgment, not passive acceptance. The test over the next year is whether that ambition becomes measurable classroom practice—and whether students gain the confidence to challenge the very systems sponsoring the lesson.