Safe and effective use of generative AI in education: what the DfE guidance means for your school

AI has moved from the early-adopter margins into everyday school work. Official DfE data for 2024–25 shows 44% of teachers using generative AI in education, mainly for planning and preparation. Twinkl’s 2025 survey puts that closer to 60%, and recent Teacher Tapp polling suggests around one in five teachers planned their most recent lesson with AI. Interestingly, the governance situation has barely shifted: most teachers report no meaningful training, about half of schools still have no AI policy at all, and two-thirds lack any rules specific to pupil use.

This article is designed as a leadership-level guide to the current DfE landscape on generative AI in education: what’s been published, what’s changed and evolved in the updates, and how to act on it safely in your setting.

AI Tutoring Policy for Schools

A practical framework to help you integrate AI tutoring in your school and meet data protection and safeguarding requirements – includes an editable policy template

Download Free Now!

What the DfE has published: the latest picture

The DfE’s generative AI guidance for schools now rests on three main pillars:

1: The Generative AI product safety standards

The product safety document has been renamed and expanded to Generative AI: product safety standards, updated on 19 January 2026 and launched at the UK AI for Education Summit.

2: The Using AI in education settings: support materials collection

The Using AI in education settings: support materials collection was refreshed on 19 May 2026 for the 2026–27 academic year, with four modules plus a leaders’ support pack and Chartered College of Teaching assessment and webinars. This includes operational aspects, and SEND-focused materials.

3: The core policy paper Generative artificial intelligence (AI) in education.

The core policy paper Generative artificial intelligence (AI) in education remains the narrative backdrop; its first publication date is June 2025, and it is not yet clear whether it has been substantively rewritten in 2026.

You can think of these as three layers: what safe AI tools must do, how staff should be trained and supported, and the broader policy story for generative AI in education.

Generative AI tools: product safety standards

The DfE product safety standards are now the clearest statement of what “safe and effective” means for AI tools used with pupils.

They require AI tools to:

• State their purpose and evidence claims honestly, picking from defined educational use cases.
• Provide filtering and monitoring of learner-facing content, with logging, DSL alerting and clear safeguarding-disclosure routing.
• Meet security and privacy expectations: UK GDPR, Data Protection Act 2018, ICO Children’s code, and no use of pupil or teacher data or IP for commercial purposes or model training without a lawful basis or consent.

The January 2026 update added four new harm areas:

• Cognitive development: AI tools should not encourage cogntitve offloading by simply giving final answers by default; they should use progressive disclosure, hints first, prompt for pupil input, and guard against cognitive deskilling.
• Emotional and social development: AI tools should avoid anthropomorphism and “I” statements, avoid isolating language, and set time limits.
• Mental health: AI tools must detect distress signals and route concerns to human help via the DSL.
• Manipulation: no sycophancy, no deceptive or dark-pattern UX, and no engagement-maximising design intended to keep pupils in the AI tool.

For school leaders, this standard is now the checklist to use when you assess any AI product: if a vendor cannot show how they meet these requirements, the risk is likely too high. It’s worth noting that many ‘out of the box’ consumer AI tools like ChatGPT and Claude do not align with these standards, simply because they lack school-specific logging, cognitive development controls, and DSL-linked safeguarding flows.

The support – materials collection for staff and leaders

The refreshed support-materials collection is structured as four modules plus a separate leaders’ pack:

• Module 1 – Understanding AI in education
• Module 2 – Interacting with generative AI in education
• Module 3 – Safe use of generative AI in education (the DfE explicitly says this should be used regardless of experience level)
• Module 4 – Use cases of generative AI in education

Alongside these sits Using AI in education: support for school and college leaders, and the Chartered College adds a certificated multiple-choice assessment and webinars mapped to the modules.

This gives you a ready-made structure: staff literacy, safe use, and practical use-cases, underpinned by a leader-specific pack you can use to help staff explore different generative AI tools and their practical applications in educational settings as part of your whole-school approach. All at no cost!

Safeguarding context: KCSIE and Ofsted

The standards explicitly cite Keeping children safe in education (KCSIE) 2025 parts 1, 2 and 5. KCSIE 2025, in force from 1 September 2025, addresses generative AI within online safety. KCSIE 2026 is due in September 2026, so any AI policy you develop now should be designed to align with the 2025 wording but flexible enough to absorb the 2026 refresh.

There is currently no AI-specific Ofsted publication: the September 2025 inspection toolkit makes no mention of AI, and the December 2025 Annual Report is themed on disadvantage, not AI. Let’s wait and see if Ofsted-shaped guidance is published in due course.

Why the guidance matters now

The guidance has evolved at an important moment, where AI is becoming more present in day-to-day school life for many staff and pupils. Here’s a round-up of how schools are usng AI:

On the staff side:

• In the DfE Technology in schools survey 2024–25, 44% of teachers reported using generative AI for school activities, most commonly for lesson planning, with much lower use for live lessons and marking.
• Independent surveys suggest the trend is sharper still: a 2025 Twinkl survey found 60% of UK educators using generative AI in education (up from 30% in 2023), and an NEU 2026 survey found 76% of teachers now using AI day-to-day.

On the pupil side, the picture is of AI-native learners whose habits are ahead of many schools’ policies:

• Ofcom’s 2025 Children and Parents Media Use and Attitudes report found that 50% of children aged 8–17 had used generative AI in 2025, up from 46% the year before, with educational use rising fastest; among AI-users, “to learn” and “for school” each rose to 45%.
• Oxford University Press’s Teaching the AI-Native Generation study found that 80% of UK students aged 13–18 used AI in their schoolwork in 2025, yet only 47% felt confident identifying accurate AI-generated information, and 51% worried resources may be biased or reinforce untrue stereotypes.
• National Literacy Trust research with 60,000 young people found two-thirds of 13–18s using generative AI in 2025, with homework help among the top uses, and weekly use rising strongly year-on-year.

Parents and teachers are alive to the risks: Parentkind reports that only 12% of parents think AI should be used in lessons, with concern focused on over-reliance and loss of skills, and 60% of pupils worry AI encourages copying over original work. NEU data shows 66% of secondary teachers think pupils’ critical thinking has declined because of AI. The government’s plan for piloting AI tutors for disadvantaged pupils has also been met with a fairly muted reception.

Against this backdrop, the DfE’s standards and support-materials are not optional extras; they are the baseline for any school that wants to meet its safeguarding, curriculum and equity responsibilities while adopting AI safely.

Five themes: our reading of the guidance

The original article organised the early guidance into five themes and a 10-step leadership guide. Those themes and steps were our editorial synthesis, not the DfE’s own structure. In this refresh, we keep the synthesis because it helps leaders act, but we are explicit that it is Third Space Learning’s reading of the guidance, sitting alongside the formal modules and standards.

1. Safety, data privacy, and responsibility

Safety now means more than filters. It means matching generative ai technologies and generative ai tools to clearly defined educational purposes, meeting data-protection obligations, addressing data privacy and intellectual property concerns, and aligning with KCSIE and the four harm areas in the product safety standards.

This is also where the sharpest new risks sit: the rapid rise of AI-generated child sexual abuse material online, and evidence from schools of pupils creating sexually explicit deepfakes.

2. Ethics and fairness

Ethical use of generative AI in education means confronting bias, hallucination and equity head-on, not assuming the technology is neutral.

• Many young people worry AI tools and resources may be biased or reinforce untrue stereotypes.
• Data on children’s AI use show persistent digital divides: children in lower-income households are less likely to have used tools like ChatGPT, and overall AI use is higher among higher-income cohorts.

The government’s equity framing matters here: the Every Child Achieving and Thriving white paper is explicit that access to tutoring is deeply unequal, and that AI tutoring policy is partly about bringing one-to-one support, often reserved for the privileged, to scale.

3. Workload and efficiency

The evidence on workload is still emerging, but there are clear signals that well-scaffolded use can reduce administrative burdens for teachers without harming quality.

• An EEF/NFER trial found that teachers using ChatGPT alongside a support guide cut weekly lesson and resource planning time by around 31%, with no noticeable difference in lesson quality.
• Some things generative AI tools can do for you or your teachers:
  • Automate grading and provide personalised feedback, especially for time consuming tasks tied to reviewing student work.
  • AI can also generate quizzes and exercises for courses.
  • Generative AI can provide instant, personalised feedback to students.
• Developer-reported data cited by government suggests larger potential savings in specific AI tools, but these should be treated as pilots, not independent trials.

The guidance is clear that workload gains must not come at the expense of cognitive development or safeguarding; the standards explicitly guard against AI tools that just give answers, even where some systems use data analysis of student performance to tailor content.

4. Human-in-the-loop

Across the guidance and the white paper, one line is consistent: generative AI should complement high-quality teaching, not replace it.

• The white paper states plainly that simply using general-purpose generative AI in education often has negative outcomes, because it “simply gives answers rather than helping children learn and building their cognition”; by contrast, AI tools specifically built for education and rooted in good pedagogy show promising signs.
• It commits that AI tutoring tools for secondary pupils will be co-created with teachers and industry, available by the end of 2027, and will complement high-quality face-to-face teaching, not replace it.

For leaders, this is the frame to use in your own policies: AI should sit under teacher oversight, within curriculum and assessment design, and within safeguarding systems — not as a standalone “assistant” pupils are left to navigate alone.

5. Staff confidence and CPD

The data shows a sharp training gap:

• In 2025, most teachers reported not having had adequate training on AI, despite large numbers already using it in their work, and only a small minority felt AI was well-regulated.
• By 2026, nearly half of schools still had no AI policy for staff or students, and two-thirds had no policy specific to pupil use.

The DfE’s modules and leaders’ pack, plus Chartered College assessment, give you ready-made content for closing this gap; the challenge is prioritising time and helping staff develop understanding of AI tools, stay informed, and build confidence through a whole-school plan that supports stronger educational practices across the school and wider education sector. Third Space Learning’s free AI literacy course for school leaders is a good starting point for building this into your CPD plan.

Taking the lead: a 10-step leadership guide

You can adapt this to your context, but as a starting point:

1. Name your AI purpose – Decide where AI can genuinely help (planning, admin, specific tutoring, accessibility), including to create lesson plans or personalised materials where they can enhance engagement and motivation, and where it must not be used (assessment design, unsupervised pupil help that just gives answers). Link each use to a clear educational purpose and evidence base.
2. Map current use – Use staff surveys and pupil voice to understand which AI tools are already in use, for what, and with what safeguards; expect reality to be ahead of AI policy.
3. Make the product safety standards your procurement baseline – Require vendors to show how they meet each point, including the four harm areas and data-protection obligations.
4. Align with KCSIE 2025 and safeguarding – Ensure your AI policy and any AI tool deployment routes concerns through DSL structures and sits visibly within online-safety and safeguarding policies.
5. Use the DfE support materials as your CPD structure – Plan staff training across Modules 1–4, with Module 3 as non-negotiable, and integrate the leaders’ pack and Chartered College assessment where appropriate, including curriculum design as a practical use case.
6. Set clear classroom guidelines for pupil use – Decide how far pupils can use AI for drafting or homework, how you will guard against cognitive offloading, and how you will teach students to interrogate AI-generated content for bias and accuracy in ways that preserve their own understanding and critical thinking skills.
7. Define your human-in-the-loop rules – Make explicit that teachers remain responsible for curriculum, assessment and pastoral care; AI tools are supports, not decision-makers, echoing the white paper’s “complement not replace” line.
8. Plan for equity and the digital divide – Use data on disadvantage gaps, tutoring impact and AI access to ensure your AI strategy closes gaps rather than widening them, for example by focusing AI investments on structured support for disadvantaged pupils and using AI to enable personalised learning that adapts to student needs and different learning styles.
9. Monitor, review and retire tools – Build in processes for reviewing tools against the standards and your own outcomes, and for retiring tools that prove too risky or ineffective; track learning outcomes, whether tools support learning, and whether use is building deeper understanding rather than just speeding up task completion.
10. Communicate with parents and pupils – Share your approach transparently, acknowledging both benefits and risks, and invite questions; use evidence from parent and pupil surveys to show you take concerns seriously.

What’s new in this refresh

Since the original guidance was published, several things have evolved.x

The headline changes are:

• The product safety expectations have become a full Generative AI: product safety standards document, renamed and expanded in January 2026 with explicit cognitive, emotional/social, mental-health and manipulation harm areas.
• The support-materials collection has been refreshed for 2026–27, with a sharpened module structure and leaders’ pack, plus Chartered College assessment and webinars.
• The government’s Every Child Achieving and Thriving schools white paper (February 2026) has pivoted decisively towards AI tutoring as a scalable route to equity, confirming and extending the product-safety focus on mental, social and cognitive development across the wider education sector.
• KCSIE 2025 is now in force, with generative AI explicitly addressed within online safety, and KCSIE 2026 is on the horizon for September 2026.

Limitations and what to watch next

The data and policy picture is still moving, and there are things you should be cautious about claiming, with future research still needed on long-term impacts across the educational system.

• It is not yet clear whether the core Generative artificial intelligence (AI) in education policy paper has been substantively rewritten in 2026; avoid stating that it has without checking the GOV.UK page directly.
• Some evidence is flagged as developer-reported and should be used as indicative, not as proof; independent trials like the EEF ChatGPT study carry more weight because they are grounded more clearly in science, but still require nuance.
• There are genuine data gaps, for example on pupils using AI specifically for maths homework in 2025–26; resist the temptation to manufacture numbers to fill these gaps.

Looking ahead, key things to watch include:

• The EEF/NFER trial of Oak’s Aila lesson-planning assistant, with a report due in autumn 2026.
• The Curriculum and Assessment Review recommendations on AI and assessment, including explicitly building AI into Computing and leaning towards exams over coursework, with a revised curriculum due in 2027 for first teaching from 2028, as schools also watch how large language models and natural language processing continue to shape academic writing and assessment guidance.
• The publication of KCSIE 2026 and any future Ofsted guidance that addresses AI more directly.

AI tutor Skye: an example worth noting

Within this policy landscape, AI maths tutor Skye is one example of generative AI in education with real life practical applications, designed to meet the DfE’s product safety expectations.

Skye’s design starts from two non-negotiable distinctions:

• Skye delivers lessons designed by expert teachers; AI is simply the delivery mechanism, not the curriculum creator.
• Skye gives hints, not answers: the progressive-disclosure principle the product safety standards now expect for cognitive development, helping students build knowledge and skills across various domains in maths while maintaining their ability to reason independently.

From a safety standpoint:

• Sessions are audio-only, with a shared whiteboard of pre-approved, teacher-made lessons; Skye has no access to the wider internet and cannot generate anything beyond the lessons.
• All sessions are recorded and transcribed, with school staff supervising; concerns are flagged, not “detected”, and escalated to the school’s DSL, aligning with KCSIE and the mental-health and safeguarding expectations in the standards.
• Skye is hosted on AWS in the UK/EU, aligned with UK GDPR and the Data Protection Act 2018; Third Space Learning is the data processor, and the school remains the data controller.

On the emotional/social and manipulation fronts:

• Skye’s conversational design avoids anthropomorphism and “I” statements and is time-bound, with finite lessons rather than engagement-maximising loops.
• There are no dark-pattern or attention-extending UX features; the aim is to help pupils work through a maths lesson, not to keep them inside the programme.

In terms of impact and credibility:

• In its first year, Skye supported hundreds of pupils across dozens of schools, delivering thousands of one-to-one sessions. Within-session quiz correctness rose substantially, and a clear majority of pupils reported a confidence increase.
• A high proportion of tutored pupils in surveyed schools reached the expected standard compared to the national average. An independent case study of Skye by Educate Ventures looks at this evidence in more depth.

Where to begin

If you are at the beginning of your school’s AI journey, or trying to move from ad-hoc use to something more deliberate, the most practical place to start is with:

• The Generative AI: product safety standards as your procurement and AI tool-assessment checklist, including how you evaluate different generative AI tools and whether large language models are being used appropriately.
• The Using AI in education settings: support materials modules and leaders’ pack as your CPD structure, particularly Module 3 on safe use, with practical examples of how staff can create resources and how Generative AI can analyse student performance data for tailored content.
• A simple, honest map of where AI is already being used in your school, and how that lines up against the standards, safeguarding and your curriculum priorities, so you can explore create-focused use cases while keeping human oversight in place.

Third Space Learning’s free AI literacy course and wider resources on AI policy for schools, AI training for teachers and AI tutoring evidence can help you use this guidance in practice, rather than starting from a blank page.