The DfE says use AI responsibly. What does that mean in practice?

The Department for Education's guidance on generative AI in education is sensible. It leans hard on professional judgement, data protection, and child safety. All things teachers already take seriously. (I'm referring to the DfE guidance last updated 12 August 2025.) (GOV.UK)

But when I talk to teachers about AI marking, I keep hearing the same question:

What does "responsible" actually look like when I'm sat at my desk on a Tuesday evening with 30 essays to mark?

The guidance gives us principles. What it doesn't give us is a playbook.

So here's mine:

• Five practical questions your AI-marking workflow needs to answer
• Reasonable defaults (what "good judgement" tends to look like in practice)
• A one-page internal policy template you can copy and adapt
• And a useful extra: the DfE's Generative AI product safety standards (updated 19 January 2026). This is the closest thing we have to "this is what safe tools should actually do". (GOV.UK)

---

What the guidance actually says (in one line)

The core message is simple:

"Teachers, leaders and staff must use their professional judgement…" (GOV.UK)

And the DfE is clear that AI can speed things up, but it cannot replace subject expertise and teacher judgement. (GOV.UK)

All fair. The problem is the bit that's left to schools and departments to work out. What counts as "responsible" when you're doing assessment at scale?

---

The five questions the guidance doesn't answer (but your workflow must)

1) What does "review" actually look like?

The DfE says AI output needs critical judgement and checking. (GOV.UK)
In practice, teachers need a standard that's workable.

A sensible default:

• For summative / reportable assessment: you read the feedback you're about to release for every student, and you sanity-check the mark against the response.
• For low-stakes drafting: sampling can be fine, but you are clear with students it's draft-level support, not "the mark".

If you wouldn't feel comfortable defending it in a corridor conversation with your HoD, it probably hasn't been reviewed enough.

---

2) How do you maintain an audit trail?

If a parent asks how their child's essay was marked, what can you actually show?

A sensible default: keep three things together:

• the question
• the student response (ideally anonymised inside the tool)
• the final teacher-approved feedback and mark

Better still: keep the AI draft vs teacher-edited final (even if that's just a simple note like "AI suggested 14/25; teacher final 12/25. Evaluation not sustained."). That protects staff, and it helps departments moderate.

---

3) What do you tell parents?

The DfE explicitly nudges schools to think about parent engagement around AI use. (GOV.UK)
Most schools are still making this up as they go along.

A sensible default message (plain English):
"We may use AI tools to help generate draft feedback, but a teacher checks and approves all feedback before it's shared. Teachers remain responsible for final marks and comments."

You're not asking parents to "like AI". You're explaining the workflow and reassuring them where accountability sits.

---

4) How do you ensure consistency across a class?

When you mark 30 essays yourself, you're the constant. Even if you're tired, you're still "consistently you".

If you're pasting answers into a general chatbot one-by-one, you lose the cohort view. And without that, you can't easily spot:

• drift (AI being harsher on some scripts than others)
• weird outliers
• a prompt that's accidentally shifted your standard mid-session

A sensible default: you need some kind of class-level view. Even a simple moderation step helps: scan the marks distribution and check a few scripts at the top, middle and bottom before you release.

---

5) What's the policy for unsupervised homework?

The DfE suggests schools may want to review homework and unsupervised study in light of generative AI. (GOV.UK)

This is not just about cheating. It's about what homework is for now that AI exists.

A sensible default: separate homework into two buckets:

• Practice that's meant to build independent skill (AI use restricted, disclosed, or tightly scaffolded)
• Work that's meant to produce a polished output (AI use permitted, with disclosure rules)

If a student uses AI to write it and you use AI to mark it, you can end up with a very polished loop where nobody's thinking especially hard. That's not a safeguarding issue. It's a learning design issue.

---

The missing piece: the DfE's "product safety standards"

In January 2026 the DfE updated its Generative AI: product safety standards. It's aimed at suppliers, but it's genuinely useful for schools because it describes what safer education AI products should actually do. (GOV.UK)

A few standards that matter immediately:

• Clear stated purpose and use cases (so you know what the tool is for, and what it is not). (GOV.UK)
• Filtering that reliably prevents harmful or inappropriate content, maintained across the whole conversation, and adjusted for age and SEND needs. (GOV.UK)
• Monitoring and reporting, including robust activity logging (recording prompts and responses) and alert routes when risks appear. (GOV.UK)
• Privacy and data protection expectations (including clarity on what data is collected, how it's used, retention, and where it's processed). (GOV.UK)

This is the practical bridge between "be responsible" and "here are the features responsible tools should have".

---

What "responsible AI marking" looks like (my take)

AI proposes, teacher decides

AI can draft marks and feedback. A teacher approves the final version. The final judgement belongs to the teacher. That's exactly the DfE's framing. (GOV.UK)

There's an audit trail

If you ever need to explain or moderate a mark, you can show your working. This protects staff, and it makes moderation easier.

Data stays where it should

The DfE is clear: protect personal data, and avoid putting personal data into generative AI tools. (GOV.UK)
In practice: anonymise where you can, and use tools with clear, school-appropriate data handling.

You have a class-level view

Responsible marking is not just "is this feedback polite?". It's also: are we being fair and consistent across the cohort?

Feedback is actionable

The point of feedback isn't to justify a mark. It's to help the student improve next time. It should stay tied to the question and the assessment objectives.

---

A one-page internal policy template (copy and adapt)

AI in Assessment: Internal Policy (Draft)

1) Approved tools

• Approved tools for assessment are:
  • [List approved tools]
• Staff should not use unapproved tools for tasks involving student work or identifiable information.

2) Teacher-facing use (marking and feedback)

• AI may be used to generate draft feedback and provisional marks.
• All AI output must be reviewed by a teacher before sharing with students. (GOV.UK)
• Teachers remain responsible for the accuracy and appropriateness of all final feedback and marks.

3) Student-facing use

• Students may use AI tools only where:

  • [state what is permitted]
  • [state disclosure expectations]

• Homework categories:

  • Skill-building practice (AI use: [restricted / disclosed])
  • Polished outputs (AI use: [permitted with disclosure])

4) Data protection and privacy

• No personal data (names, identifiers) should be entered into general-purpose AI tools. (GOV.UK)
• Any approved product must have clear privacy information (what data is collected, how it's used, retention, and where it's processed). (GOV.UK)

5) Safeguarding and safety standards

• Where tools are learner-facing, the school expects alignment with the DfE's Generative AI product safety standards, including:
  • effective filtering
  • monitoring and reporting, and appropriate logging
  • clear alert routes to the DSL (GOV.UK)

6) Audit trail and moderation

• For any assessed work supported by AI drafting, the school will retain:
  • the question or task
  • the final teacher-approved feedback and mark
  • [optionally] the AI draft vs teacher-edited final (recommended for moderation)

7) Parent communication

• Parents will be informed that AI tools may be used to support (not replace) teacher marking. (GOV.UK)
• Suggested wording:
  "We may use AI-assisted tools to help teachers produce faster, more detailed draft feedback. A teacher reviews all feedback before it is shared."

8) Review

• This policy will be reviewed [termly / annually], alongside updates to DfE guidance and product standards. (GOV.UK)

---

The DfE got the principles right, but schools still need a playbook

I don't think the guidance is bad. It's deliberately broad. "Use your professional judgement" only helps if schools translate it into:

• clear workflows
• clear accountability
• clear data rules
• clear communication

The departments I see getting this right aren't waiting for perfect guidance. They're building sensible habits now.

That's the design philosophy behind Teach Edge: AI drafts quickly, but teachers stay in control, with an audit trail and a class view built into the workflow.