DfE Generative AI Guidance for Schools: What It Says and What It Means for Filtering
The Department for Education has built up a small family of documents on generative AI in schools and colleges. Read together, they say AI can help, safety comes first, and filtering and monitoring must reach AI tools too. This page explains each document and turns them into decisions your filter can enforce.
DfE generative AI guidance: the timeline
The guidance has been updated several times, so older summaries can be misleading. These are the dates from the DfE's own publication histories, which are worth checking again before each annual review.
In 2026 the filtering and monitoring standard was also updated twice on AI: in June to clarify how to assess generative AI risk, and in August to reference the generative AI product safety standards. KCSIE 2026 points to both the AI guidance and the product safety guidance. See KCSIE 2026 changes.
Decide which AI tools reach which year groups
Our hosted CIPA filter has an AI / LLM Tools category, so a school can treat chatbots as a group instead of one site at a time. Because each school has its own switches, a sixth form can have a different answer from a primary.
- Switch the whole AI / LLM Tools category on or off per school
- Use always-allow for the one approved tool and always-block for the one that worries you
- Reports show what was requested, which feeds the review the DfE expects
Generative AI in education: the DfE position in plain terms
The core guidance is aimed at schools and colleges, with a focus on compulsory education. It sees real benefits, especially for staff workload, but treats pupil-facing use as higher risk than teacher-facing use.
What schools should do
- Make safety the top priority and weigh benefits against risks before using AI.
- Be clear about the intended use and choose suitable use cases, such as staff-only or administrative use.
- Include unauthorised use in risk assessments, not just approved use.
- Only allow pupil use with safeguards such as close supervision and filtering and monitoring.
- Review homework and unsupervised study policies, and consider engaging parents.
- Check AI outputs: professionals stay responsible for what they produce.
- Comply with the age restrictions set by AI tools.
What schools should avoid
- Entering personal data into generative AI tools.
- Using pupils' work to train AI without permission from the pupil, or a parent if the pupil is a minor, unless an exception applies.
- Assuming free tools keep inputs private: many use inputs for training.
- Letting AI replace the relationship between teacher and pupil.
- Treating AI as outside safeguarding: online safety, including AI, belongs in safeguarding policies.
- Ignoring cyber security: AI can make attacks more convincing.
Assessment
Schools and awarding organisations should take reasonable steps to prevent AI-related malpractice, following JCQ and Ofqual guidance.
Ofsted
Inspectors do not evaluate AI tools directly. They look at how leaders made decisions about AI and the impact on pupils.
Intellectual property
Pupils' original work and teachers' lesson plans may be protected by copyright, which affects what can be uploaded.
Transparency
Pupils and parents should be told when their personal data is processed using AI.
The guidance is positive about teacher-facing uses such as planning, resource creation and feedback, which it sees as lower risk and a way to reduce workload. That is why many schools start by allowing AI for staff and adding pupil access later, tool by tool.
For your filter, this means staff and pupil profiles should treat AI differently. Staff profiles can allow more AI services, with training on personal data, while pupil profiles allow only tools that have been approved for their age.
Generative AI product safety standards, standard by standard
First published in January 2025 under the title product safety expectations, the document now appears as product safety standards. It is written mainly for developers and suppliers, but schools can use it as a buying checklist. Some standards apply to learner-facing products only.
Stated purpose
Products should say clearly what they are for, including target ages and subjects, without overstating what they can do.
Educational use cases
Suppliers identify the use case, such as personalised learning or assessment, and show evidence the product meets it.
Filtering
Harmful content should be blocked reliably through a whole conversation, adjusted for age and needs, across images, languages and misspellings, on any device.
Monitoring and reporting
Prompts and responses should be logged, users told when content is blocked, and the DSL alerted to possible safeguarding issues within an agreed time.
Security
Products should resist jailbreaking, let administrators set permissions, use strong authentication and fit the cyber security standards.
Privacy and data protection
A clear, age-appropriate privacy notice, a DPIA across the product's life, and no commercial use of personal data, including training, without a lawful basis.
Intellectual property
No use of pupils' or teachers' work for commercial purposes, including training, without permission.
Design and testing
Child safety built in from the start, with testing involving realistic users, including children, before each release.
Governance
A risk assessment for educational use, a formal complaints route and understandable AI safety policies.
Cognitive development
Avoid giving final answers by default, use hints and learner attempts first, and report on over-reliance.
Emotional and social development
Avoid human-like personas, do not encourage secrecy or isolation, and set default time limits only teachers can override.
Mental health
Detect signs of distress, signpost human help and raise a safeguarding flag where needed, following a published crisis protocol.
Manipulation
No flattery, false confidence, pressure, dark patterns or design that stretches engagement or steers users to paid options.
Standard numbers here follow the order on the DfE page at the time of writing. The DfE does not certify products against the standards, so ask suppliers for evidence rather than a badge.
Product filtering versus school filtering
The product safety standards ask AI tools to filter their own output. The filtering and monitoring standards ask schools to filter what pupils can reach. These are two different layers, and you need both.
| Question | Inside the AI product | School web filter |
|---|---|---|
| What does it control? | What the tool says or generates in a conversation. | Which AI websites and services pupils can reach at all. |
| Who sets it up? | The supplier, with settings for the school. | The school, its IT support or its filtering provider. |
| What does it miss? | Every unapproved AI tool, because it only protects its own product. | Harmful output inside an approved tool. |
| Who gets alerts? | The DSL, under the monitoring and reporting standard. | The DSL, through monitoring reports and filter logs. |
| Which standard? | Generative AI product safety standards. | DfE filtering and monitoring standards and KCSIE. |
Website category filtering is one layer. It controls access to sites and services, but it does not read conversations, monitor messages or watch pupil behaviour. Concerns raised by any tool go to the DSL through your safeguarding process.
What the DfE AI guidance means for filtering AI tools
Here is one way to turn the guidance into filter decisions. It is an example to adapt, not a DfE template, and your own risk assessment should drive the final settings.
| Type of AI tool | Primary pupils | Secondary pupils | Staff | Reasoning |
|---|---|---|---|---|
| Approved learning tool that meets the product safety standards | Supervised only | Allow | Allow | DfE expects pupil use only with safeguards; age limits apply. |
| General-purpose AI chat services | Block | Decide by age limits | Allow, no personal data | Many set minimum ages; staff must not enter personal data. |
| AI companion and character chat | Block | Block | Review | KCSIE 2026 names AI that simulates interaction as a contact risk. |
| AI image and video generators | Block | Approved tools only | Allow | Risk of sexualised or harmful images, including AI-generated explicit images. |
| "Undress" and deepfake tools | Block | Block | Block | No legitimate school use; serious safeguarding and legal risk. |
| Essay writers and homework answer tools | Block | Block in exams | Allow | Malpractice risk under assessment guidance; homework policy decides the rest. |
| AI detection tools | n/a | n/a | Use with care | Results are not proof; follow assessment body guidance. |
AI tools blocklist
11,800+ AI-tool domains in 18 categories, updated daily, to build these rules.
AI tools blocklist for schoolsDeepfakes and NSFW AI
Filtering the highest-risk AI content categories.
Deepfake and NSFW AI filteringAllow or not?
The arguments for and against pupil access to AI chat.
Should AI chatbots be allowed in schoolsHow well-known AI services are labelled
Some are plain chatbots, some are companions, and some are something else with AI added. The labels below are from our database.
| Domain | Categories in our data | Which kind of AI tool |
|---|---|---|
| chatgpt.com | AI / LLM Tools | General chatbot |
| character.ai | AI / LLM Tools, Web-based Applications | Character chat, which a DSL may look at closely |
| replika.com | AI / LLM Tools, Community Sites, Social Networking | An AI companion with social features |
| midjourney.com | AI / LLM Tools | Image generator |
| photomath.com | AI / LLM Tools, Education/Reference | A homework helper |
| claude.ai | AI / LLM Tools, Web-based Applications | General chatbot, listed with web applications too |
Five steps to bring AI into your filtering and monitoring review
The filtering and monitoring standard asks senior leaders to include generative AI when scoping requirements, and asks reviews to consider AI-generated content and to be triggered when new AI tools are introduced.
Discover
Use filter reports to see which AI services pupils and staff already reach.
Decide
Approve tools by age and purpose, checking age limits and supplier evidence.
Configure
Allow approved tools, block high-risk categories, and set staff rules separately.
Monitor
Agree how alerts from AI tools and the filter reach the DSL.
Record
Note how AI products are used and review when a new tool is introduced.
Put these steps into your yearly review using our filtering and monitoring annual review checklist, and read the DfE filtering and monitoring standards guide for the wider context.
Why a pupil cannot just switch to a proxy
VPN/Proxy/Anonymizers is a category the CIPA preset switches on, so a blocked chatbot is not reopened through a proxy site. Public encrypted DNS and browser settings that skip the filter are blocked by default, and a private encrypted link carries the school's rule onto laptops used at home.
Questions to ask an AI supplier, based on the product safety standards
Use these when a department asks to adopt a new AI tool. Ask for written answers, keep them with your review record, and involve your data protection officer and DSL before the tool reaches pupils.
Safety and safeguarding
- What ages and needs is the product designed for?
- How does content filtering work across a whole conversation, including images?
- What is logged, and how does the DSL receive alerts?
- How are distress signals and self-harm mentions handled?
- Can teachers set time limits and control solution modes?
Data, security and governance
- Are pupil or staff inputs used to train models? Can that be switched off?
- Where is data stored and processed, and under which legal basis?
- Is there a DPIA we can review, and a privacy notice suitable for pupils?
- How is the product tested against jailbreaking?
- What is the complaints and escalation route for safety issues?
Where AI appears across UK school guidance
AI now runs through several documents. This table shows where each one mentions it, so your policies point to the right source.
| Document | What it says about AI | Our guide |
|---|---|---|
| KCSIE 2026 | Contact and conduct risk definitions include AI; links to the AI guidance, staff materials and product safety guidance. | KCSIE 2026 changes |
| Filtering and monitoring standard | Scope includes generative AI; reviews consider AI-generated content and new AI tools. | Filtering standards |
| IT support standard | Training and guidance should be updated when new technologies such as generative AI tools are introduced. | Digital and technology standards |
| Online Safety Act | Duties on in-scope online services, including age assurance where required. | Online Safety Act for schools |
What the DfE AI guidance does not say
No approved product list
The DfE does not publish a list of approved AI tools or certify products. Schools judge suppliers against the standards and record their reasoning.
No blanket ban
Nothing in the guidance requires schools to block all AI. It asks for safety first, clear use cases and safeguards for pupil use.
No duty to detect AI writing
The guidance points to assessment bodies on malpractice. It does not require detection software, and detection results are not proof on their own.
No separate AI policy required
AI can sit inside existing safeguarding, online safety, data protection and assessment policies, as long as it is covered clearly.
The guidance applies to England. Schools in Wales, Scotland and Northern Ireland should check their own government's advice, though the product safety standards are a useful buying checklist anywhere in the UK.
Putting the five AI review steps into practice with the hosted filter
Each row matches one of the steps above with something the filter does.
| Review step | Without a category filter | Hosted CIPA filter |
|---|---|---|
| Discover | Ask staff and hope the list is complete | Activity reports show which sites were requested |
| Decide | One rule for every pupil | Separate settings per school, so age ranges can differ |
| Configure | Add each tool by hand | One switch for AI / LLM Tools, then always-allow for approved ones |
| New tools next term | Back to the list | 11,800+ chatbots, companions and generators we have recognised |
| Away from school | Rules stop at the gate | A private encrypted link keeps the same rules on devices |
| Record | Notes in a document | A policy record and a log export |
DfE generative AI guidance: frequently asked questions
Does the DfE allow pupils to use generative AI in school?
The DfE guidance does not ban it. It says safety should be the top priority, that benefits should be weighed against risks, and that pupils should only use generative AI with appropriate safeguards in place, such as close supervision and filtering and monitoring. Schools must also respect the age restrictions set by AI tools.
What are the DfE generative AI product safety standards?
They describe the capabilities and features generative AI products should have to be considered safe in education. They were first published in January 2025 and updated in January 2026 with new standards on cognitive development, emotional and social development, mental health and manipulation. They are aimed mainly at developers and suppliers, but schools can use them when choosing tools.
Do filtering and monitoring rules apply to AI tools?
Yes. The DfE guidance says filtering and monitoring should cover generative AI, in line with the DfE filtering and monitoring standards, and KCSIE links to the product safety guidance that explains how filtering and monitoring apply to AI in education.
Can staff put pupil data into AI tools?
The DfE guidance says personal data should not be entered into generative AI tools. Where personal data has to be used, settings must protect it and follow data protection law, and be open with pupils and parents about it.
Should schools block all AI tools?
A blanket block is rarely needed. Most schools approve a short list of tools for particular ages and uses, block higher-risk groups such as AI companion chat and image generators for pupils, and monitor approved tools. Decisions should be recorded and reviewed as part of the filtering and monitoring review.
Price depends on how many schools and devices you have, and our quote follows within one business day. A pilot in one school lets you see how the AI rules look in the blocks and reports before you take them wider.
Talk to us about your school or trust
We can show how our AI tools categories help you allow approved tools, block high-risk ones by age group and keep your rules current as new AI services launch.