The Department for Education just released its Generative AI: product safety standards, and filtering sits at the foundation. With 8 in 10 UK teenagers now using AI tools in their schoolwork (92% in London), these standards for safe AI use for students couldn't have come at a better time.

We've spent the last year working with UK schools across the country, and filtering has consistently been one of the first concerns raised by teachers and safeguarding leads. The new guidance finally gives the sector clear expectations on what safe AI in classrooms actually looks like in practice.

Why Filtering AI in Classrooms is Different

Traditional web filtering blocks access to websites based on categories and keywords. Filtering AI in classrooms is fundamentally different because it needs to understand context, intent, and the full arc of a conversation.

Consider a Year 10 history lesson on propaganda. Students might need to discuss sensitive topics around WWII, extremism, or political manipulation. A traditional keyword filter would block this entirely. But safe AI use for students in UK schools requires filtering that understands this is legitimate pedagogical work within a specific curriculum context.

The Department for Education guidance recognises this complexity. The Generative AI: product safety standards require that filtering:

Understands context throughout entire conversations, not just individual prompts
Works across all modalities: text, images, misspellings, abbreviations, multiple languages
Adjusts for age and individual needs, including SEND students
Functions everywhere: school devices, BYOD, smartphones
Remains embedded in the product itself, not bolted on as an afterthought

What This Means in Practice for Safe AI Use for Students

When we built Willow, one of the first things teachers told us was that they needed filtering that understood pedagogical intent. They didn't want sensitive topics completely off-limits—they wanted smart boundaries that knew the difference between a history essay on the Holocaust and a student attempting to access harmful content.

This is harder than it sounds. Students are remarkably creative at testing boundaries. They'll use euphemisms, images, coded language, even switch languages mid-conversation. The guidance is explicit: filters must catch all of it, consistently, throughout the entire interaction.

Age-Appropriate Filtering for UK Schools

A 7-year-old and a 17-year-old need different levels of protection and different learning opportunities. The guidance requires that filtering adjust accordingly, including for students with SEND who may need additional support or different approaches to content moderation.

This means filtering systems need to know not just what content is being discussed, but who is discussing it and in what context. A conversation about relationships in a Year 11 PSHE lesson requires different filtering than the same conversation with a Year 3 student.

The Reality Check: Student Concerns About Safe AI Use

Recent research from Oxford University Press found that fewer than half of UK pupils feel confident they can tell what information provided by AI is actually true. Even more concerning, 51% worry the AI tools they're using may not be safe.

Students know they need protection, but they also know they can't avoid AI entirely. They're using it at home, on their phones, often with tools that have no filtering whatsoever. The answer isn't to block AI in classrooms—it's to provide tools that are safe by design and meet the Department for Education's standards.

What UK Schools Should Ask of AI Tools

When evaluating AI tools for use in classrooms, the new Department for Education standards give you a clear framework for questions to ask:

On contextual filtering:

How does your product understand the difference between legitimate educational discussion of sensitive topics and harmful content?
Can you demonstrate how filtering works across a multi-turn conversation?

On multimodal protection:

How do you filter content across text, images, and other formats?
What happens if a student tries to circumvent filters using misspellings, slang, or other languages?

On age-appropriate filtering:

How does your product adjust filtering for different age groups?
Can it accommodate individual needs, particularly for SEND students?

On embedded protection:

Is filtering built into your product's core architecture, or is it an add-on?
Does filtering work consistently regardless of device (school devices, BYOD, smartphones)?

Building for Safe AI Use for Students from Day One

The requirement that filtering be "embedded within products" is significant. It means edtech companies need to build filtering as a core part of their architecture from the beginning, not as something added later when a school raises concerns.

This is how we approached Willow. Filtering isn't a separate layer bolted on—it's woven throughout the system. Every interaction a student has is contextually filtered based on their age, the curriculum area they're working in, and the specific learning objective they're pursuing.

This approach means teachers don't need to worry about gaps in protection. Whether a student accesses Willow from a school laptop in the classroom or their phone at home, the same level of contextual filtering applies.

The Balance: Protection Without Stifling Learning

The most challenging aspect of filtering in education is maintaining this balance: protecting students from genuine harm while still enabling rich, complex learning opportunities.

A well-designed filtering system should be invisible when learning is happening as intended. Students shouldn't hit unnecessary barriers when exploring challenging historical topics, scientific concepts, or social issues that are part of their curriculum. But it should be immediately present when someone attempts to access genuinely harmful content.

This is where contextual understanding becomes critical. The same words or concepts that are appropriate in one context (a biology lesson, a history essay) might be inappropriate in another. Effective filtering understands this distinction and applies protection intelligently.

What Happens Next for UK Schools

The Department for Education's Generative AI: product safety standards represent a significant step forward for safe AI use for students. For the first time, UK schools and edtech companies have clear guidance on what safe AI in classrooms means in educational settings.

For schools currently using AI tools, this is the moment to have conversations with your vendors about how they meet these standards. For UK schools considering AI adoption, these standards give you a clear framework for evaluation.

The next pillar we'll explore is monitoring and reporting—how schools actually know what's happening when students use these tools, and what safeguarding measures need to be in place.

About Willow

Willow is an AI platform built specifically for education, giving every student a personalised teaching assistant that is grounded in strong pedagogy, aligned to your curriculum, and fully under teacher control. Safety by design is at the core of everything we build.