When AI Enters the Classroom, Where Do the Guardrails Live?
I keep coming back to the same question after reading a run of recent pieces on AI in education: where do the guardrails actually live?
According to Education Weeks survey coverage, 82% of teachers said they had not received formal guidance on how to apply AI to their work. The same report says 58% had no guidance on using AI for grading and feedback, and 69% had no guidance for one-on-one instruction or tutoring. At the same time, EdWeek notes that AI use among teachers has climbed sharply since 2023.
That is the part that feels unfinished to me. The tools are moving into everyday use, while the rules are still catching up.
A Tes guide for teachers using AI makes the practical side of this very clear. It starts with the basics, secure tools, approved policies, anonymised data, careful checking, and keeping a human in the loop. That is sensible advice, but it also reveals how much of the burden still sits with individual practice. In a messy real-world setting, that means the quality of AI use can vary wildly from one classroom to the next, even inside the same institution.
The interesting software question is where the safer defaults should live.
If a tool is being used by adults to support children, then data handling cannot be an afterthought. Neither can age-appropriate design, logging, review steps, or limits on what the model is allowed to do. Those are product decisions as much as policy decisions. If they are left only to local documents, the result is often inconsistency. If they are built into the tool, the system becomes easier to use well, and harder to use badly by accident.
That is why the op-ed in Education Week matters. It argues that the burden of proof should sit with the technology and the people pushing it, rather than with the person asking hard questions about whether a tool really supports learning. I think that is the right instinct. The more powerful the tool, the more the design needs to do some of the work.
The rise of purpose-built tools also shows how the market is trying to answer this. The 74s interview with MagicSchools founder describes an AI product aimed at administrative support, which is a very different safety problem from a child-facing chatbot. And Googles AI search trends report suggests the public is still asking a basic question too, how can AI help? That sounds simple, but in education the answer depends entirely on where the risk sits.
From a platform-design perspective, that is the real issue. If AI is going to be part of the classroom ecosystem, which safeguards should be native to the tool, and which ones genuinely belong in local policy and human practice?
Wed love to hear your thoughts on this. Find us on the social channels linked at the top of the page.