One of the clearest themes in the DfE's guidance on generative AI is data protection. If you are wondering what it actually recommends — and how that translates into what staff can type into an AI tool — here is the plain-English version.

The DfE's position in brief

The DfE's guidance on generative AI in education makes a few consistent points about data. In summary:

Our summary of the DfE's generative AI guidance covers the wider document.

What that means in practice

The account matters more than the tool

A personal, free AI account is not covered by your school's data protection arrangements. A school-licensed account — such as Copilot or Gemini on your school login — usually comes with contractual protections, including commitments not to train models on your data.

Minimise what you enter

Even in approved tools, enter only what the task needs. Most tasks — planning, letters, resources — need no personal data at all.

Special category data needs extra care

Health, SEND, ethnicity, religion and safeguarding information is special category data. It should only go into AI tools where your data protection impact assessment explicitly allows it, and often not at all.

Want a straight answer for your school?

Ask me anything about AI in your school — policy, tools, training or safeguarding. I reply personally, usually the same day.

Simple rules for staff

  1. Use only AI tools on your school's approved list.
  2. Use your school account, not a personal one.
  3. Never enter pupil names or identifying details into personal accounts.
  4. Anonymise wherever you can — "Pupil A", "a Year 5 pupil".
  5. Keep special category data out unless your policy explicitly allows it.
  6. If you are unsure, ask your DPO before, not after.

These belong in your staff AI acceptable use policy.

What if data has already gone into the wrong tool?

Tell your data protection officer promptly. They will assess whether it is a personal data breach and whether it needs reporting. See data breaches from uploading pupil names to ChatGPT and is ChatGPT GDPR compliant for school data?

For leaders

Complete a DPIA for each AI tool that processes personal data, keep an approved tools list, and train staff on the rules above.

Examples: allowed or not?

ScenarioUsually OK?
Planning a lesson in Copilot on your school accountYes — no personal data
Drafting a generic letter to all parentsYes
Report comments using initials in a school-approved tool covered by a DPIAUsually, if your policy allows
Pasting a named pupil's SEND report into a personal ChatGPT accountNo
Uploading a class photo to a free AI image toolNo
Summarising anonymised meeting notes in an approved toolYes

Pupil photos and AI

Images of identifiable pupils are personal data. Uploading class photos to AI tools — for example, to create cartoon versions for a display — needs a lawful basis, consent that covers it, an approved tool, and usually a DPIA. In most cases it is simpler not to. See AI-generated imagery risks in UK schools.

Transparency with families

Your privacy notice should mention how the school uses AI tools where personal data is involved. See how to explain AI use to parents.

Frequently asked questions

What does the DfE recommend regarding personal data and generative AI?

Protect personal and special category data in line with data protection law, be aware AI tools may store and learn from what is entered, do not let pupils' work train models without consent, and be transparent.

Can teachers put pupil names into AI?

Not into personal accounts. Only into school-approved tools where your policy and DPIA allow it, and anonymising is safer.

Is using ChatGPT with pupil data a GDPR breach?

Entering identifiable pupil data into a personal account not covered by school agreements can be a breach. Report it to your DPO promptly.

What is special category data in schools?

Health, SEND, ethnicity, religion and similar sensitive information. It needs extra protection and should rarely go into AI tools.

Do school Copilot and Gemini accounts protect data?

School-licensed accounts usually include contractual data protections, including commitments not to train models on your data.

← Back to all articles Ask me a question →