Most schools I know are caught between two positions on AI that both feel wrong.
The first position is policing. Detect it, ban it, treat every submission as a potential integrity breach, build the policy around what happens when a student uses it without permission. This position feels like the responsible one. It also feels increasingly like trying to ban calculators in 1975: technically possible, practically exhausting, and ultimately beside the point.
The second position is permissiveness. Let students use AI tools, trust them to be honest about it, focus on the learning not the tool. This position feels progressive. It also feels like it's skipping several steps, specifically the steps where students develop the judgement to know when AI output is useful, when it's wrong, and when it's doing their thinking for them.
Neither position has a satisfying answer to the question that actually matters: what are we trying to develop in students, and does our current approach to AI get us there?
How schools ended up here
It's worth being honest about why most schools landed in policing mode before we talk about moving out of it.
AI writing tools arrived faster than pedagogy could respond to them. A student could submit AI-generated work that looked indistinguishable from their own writing, and schools had no reliable way to identify it. The integrity risk was real and the response (tighten assessment conditions, increase supervision, treat AI as a threat) was a reasonable institutional reaction to an ambiguous situation. Detection tools have a place in that response, within limits, which I've written about in my teacher's guide to AI detection and risk reports.
The problem isn't that schools got it wrong. The problem is that the policing frame hardened into a default position before the pedagogical question was properly asked. Schools built policies around what students shouldn't do with AI before they'd worked out what students should be able to do with it.
That's the gap most schools are sitting in now. The policing policy exists. The pedagogical framework doesn't.
The question worth asking instead
The most useful reframe I've found is this: what does a student who uses AI well actually look like?
Not a student who avoids AI. Not a student who outsources their thinking to it. A student who can use it as a tool, interrogate its output, identify where it's wrong or shallow or missing the point, and produce work that reflects their own thinking rather than a plausible imitation of it.
That student has a specific set of skills. They can evaluate sources critically, including AI-generated ones. They understand that fluency isn't the same as accuracy, that confident prose can contain fundamental errors, that an AI that doesn't know your rubric will tell a Band 4 essay it's strong work. They know the difference between using a tool to think better and using a tool to avoid thinking.
Those are not new skills. They're the same evaluation skills schools have always tried to develop. AI just makes them more urgent and more visible.
What the shift from policing to partnering actually means
It doesn't mean removing guardrails. It means changing what the guardrails are for.
Policing guardrails are designed to keep AI out. Pedagogical guardrails are designed to ensure students are doing the thinking, regardless of what tools they're using. The difference isn't permissiveness. It's purpose.
In practice this looks like explicit teaching of AI literacy alongside subject literacy. Not a standalone digital citizenship unit that gets taught once and forgotten, but integrated into the way students engage with feedback, sources, and their own drafts across every subject.
It looks like assessment design that makes AI dependency visible rather than invisible. Tasks that require students to show their thinking process, not just their final product. Drafts with timestamps. Oral defences of written work. Reflection tasks that ask students to evaluate the feedback they received, including AI feedback, and explain what they did with it and why. In my own Year 11 Ancient History, that looks like a source analysis where the draft comes in with the student's notes on which sources they chose and why, then a 5-minute conversation where they talk me through the argument before the final is due. A student who handed the thinking to an AI can't walk me through it.
It looks like teachers who model critical engagement with AI output rather than either deferring to it or dismissing it. A teacher who says "MarkMate flagged your source integration as weak. Do you agree? Show me the sentence and let's look at whether it's right" is teaching something more valuable than either "don't use AI" or "here's your AI feedback, good luck."
The skill underneath all this
This is where I think the conversation in most schools hasn't gone far enough.
Teaching students to use AI well means teaching them to be genuinely critical of it. Not sceptical in a generalised way. Specifically critical. Can they identify when an AI response is confidently wrong? Can they recognise when it's given them generic feedback that doesn't actually apply to their work? Can they articulate why a piece of AI-generated writing sounds plausible but misses the argument?
These are harder skills to teach than "don't use AI" and harder to assess than "submitted without AI assistance." They're also the skills that will matter most when these students are in workplaces where AI tools are standard and the differentiating ability is knowing when to trust the output and when to push back on it.
The schools that figure out how to teach this explicitly, not as an add-on but as a core part of how students engage with feedback and sources across every subject, will be producing graduates who are genuinely better equipped than the ones who spent their school years in an AI-free zone that didn't exist anywhere else in their lives.
Where this leaves school leaders right now
If you're a Head Teacher, Deputy, or curriculum leader reading this, I'm not suggesting you dissolve your AI policy and start from scratch. The integrity risk is real and the policy exists for good reasons.
What I am suggesting is that the policy is a floor, not a ceiling. It sets the minimum: what students can't do, what constitutes a breach, what the consequences are. It doesn't set the aspiration.
The aspiration is a student who doesn't need the policy because they have the judgement to use AI well. Getting there requires a pedagogical framework that sits above the policy, not just alongside it.
That framework starts with the question: what does a student who uses AI well actually look like? Work backwards from that and the curriculum conversations, the assessment design decisions, and the professional learning priorities tend to follow.
A note on MarkMate in this context
MarkMate is designed to sit inside a pedagogical framework, not replace one. The class link feature, where students self-check drafts under teacher supervision before submission, is the use case that fits this conversation most directly. Students engage with AI feedback, evaluate it against their own understanding of the rubric, decide what to act on and what to push back on, and submit work that reflects their thinking.
That's not a policing tool. It's a thinking tool. The distinction matters.
If you're working through what a pedagogical AI framework looks like for your faculty, the Compliance & Safety page covers how MarkMate handles student data and where the boundary between AI and teacher judgement sits.
Worth sharing with your leadership team before your next AI policy conversation.
