Teaching Deepfakes and AI-Generated Content — A Guide for Schools

Why now

Seeing stopped being believing — and the curriculum caught up

The updated RSHE guidance, statutory from September 2026, names deepfakes and AI-generated imagery in the content secondary schools must cover — alongside the misinformation they power and the sexual harms they enable. That's a recognition of something teachers already know: this isn't a future risk. It's in students' feeds, group chats and, increasingly, their schools.

AI-generated content reaches teenagers three ways. As misinformation — synthetic clips and images built to be shared angry. As fraud — cloned voices and fake celebrity endorsements bolted onto ordinary scam scripts. And, the sharpest edge, as sexual harm — "nudification" tools that turn an ordinary photo into a fake nude, used between peers and by strangers. Internet Matters' research on that last category should settle any doubt about whether this belongs on the curriculum:

4 in 30

students in an average class have some experience of nude deepfakes — sending, receiving, encountering one, or a nudification app (Internet Matters, 2024)

99%

of nude deepfakes depict women and girls

2×

teenage boys are twice as likely as girls to report an experience with a nude deepfake — 18% against 9%

Scope: teaching the topic vs handling an incident

This guide is about teaching. If an AI-generated intimate image of a student surfaces — made by a peer or a stranger — that's a safeguarding matter for your DSL, handled under your school's own procedures and the UKCIS sharing-nudes guidance, which explicitly covers AI-generated imagery. For the blackmail version of that incident, see our sextortion guide for schools.

The trap

"Spot the fake" is not a strategy

The obvious lesson is the wrong one. Teaching students to hunt artefacts — mangled hands, warped jewellery, gibberish text in the background — works on today's cheap fakes and is already failing on good ones. Every artefact checklist has a shelf life, and it's measured in months: the models improve, the checklist rots, and the lesson ages with it.

Worse, artefact-hunting quietly teaches the inverse rule: if I can't see anything wrong, it must be real. That's precisely the reflex a good fake exploits — the same mistake as teaching students that a real-looking web address makes a message safe. Judging content by how it looks is the habit to break, not the skill to sharpen.

What holds up is provenance. Who posted this? Does it appear anywhere you already trust? If a clip or claim matters, don't trust the version that arrived in the feed — go and find it at the source: the official account, the official site, a named outlet. "It looks real" and "it looks fake" are both worthless evidence; where it comes from and what it wants from you are not.

Try the endpoint of that lesson

Our free activity Spot the Fake runs students through ten real-or-AI faces. The early rounds teach the classic tells; the final rounds have none worth finding — the game's own verdict is that your eyes aren't enough, so verify instead. Like all our free activities, it's browser-based and login-free, and produces no record — a taster for the classroom, not the platform.

What to teach

Six messages that survive the models improving

The technical detail will be out of date before a scheme of work finishes its first year. These won't:

"Looks real" proves nothing

And neither does "looks fake". The first job is retiring the eyeball test as evidence in either direction — a habit students find genuinely hard to drop, because it has worked their whole lives.

Source before surface

Where did this arrive from, who posted it first, and does it appear anywhere you already trust? If it matters, go direct to the official account or site rather than trusting the version that was forwarded to you.

Making one is the harm

"It's not really them" is false comfort. A fake nude of a real classmate humiliates a real person, and the law now treats making or requesting one as a crime, not a prank. This is the message with the most protective power per minute of teaching.

Scam deepfakes keep the old tells

A cloned voice or a fake celebrity endorsement is still a scam underneath: pressure, urgency and an ask — money, a code, a login. A panicked voice call asking for money? Hang up and call back on the number you already have.

Don't be the amplifier

Outrage is the distribution mechanism — "can you believe this?!" spreads a fake faster than belief does. The habit to build: verify before you forward, or don't forward.

Victims are never to blame

Being in a photo is not a mistake. And there are real routes to help: Report Remove takes reports of AI-generated intimate images of under-18s as well as real ones — around one in five of its reports now involve faked imagery.

The law

"It's only AI" — the defence that no longer exists

Students widely assume AI-generated images sit in a legal grey area — regrettable, maybe, but not really against the rules. For images of adults that was arguably true until recently. It isn't now, and KS4 students in particular need to hear how quickly it changed:

If it depicts an under-18, everything is illegal

An AI-generated sexual image of a child is an indecent image of a child in law. Making it, possessing it or sharing it is a crime — whoever made it, whatever tool made it, whether or not it's "really" them.

Creating or requesting is now an offence

Since February 2026, creating a sexually explicit image of an adult without consent — or asking an app or AI tool to create one — is a criminal offence under the Sexual Offences Act, even if it's never shared. Using a nudification tool to create one is the offence.

Sharing was already an offence

Sharing intimate images without consent — real or AI-generated — was criminalised by the Online Safety Act 2023 reforms, and platforms are required to treat it as a priority offence and proactively tackle it.

The Children's Commissioner's April 2025 report on nudification tools found them widely available and often free, concluded there is "no positive reason for these to exist" — and documented girls already changing what they post to avoid becoming raw material. The law has since caught up with the tools themselves: since 29 June 2026, making, adapting, supplying or offering to supply a nudification tool is its own criminal offence under the Crime and Policing Act 2026, carrying up to three years' imprisonment. Enforcement against overseas-hosted apps will take time to bite, though — which is exactly why the teaching still can't wait. (This section is background for the classroom, not legal advice.)

Sequencing

Where it fits at KS3 and KS4

Like every topic in a sequenced online safety curriculum, this one builds: KS3 installs the habits while the stakes are low, KS4 applies them to the harms older teens actually meet.

Key stage 3

Install the habits

What synthetic media is: images, voices and video can all be generated — and the tools are free and everywhere
Source before surface as a reflex: who posted it, and can you find it somewhere you already trust?
"Looks real" and "looks fake" both prove nothing — verification does
Don't forward what you haven't verified — sharing is publishing
The behaviour baseline: making fake images of real people is harm, not comedy

Key stage 4

Apply them to the real harms

Nude deepfakes, nudification tools and the law — creating, requesting and sharing are all offences
The sextortion overlap: the IWF reports under-18s blackmailed with nudified versions of their own ordinary photos
AI-driven fraud: cloned voices and celebrity investment deepfakes — the same underlying tells of pressure, urgency and an ask
Misinformation at scale: reading the incentive behind a viral clip before trusting it
Digital footprint, revisited: public photos are now raw material

Sources