Deepfake Cyberbullying: How It Works, How Common It Is and How to Respond

What is deepfake cyberbullying?

Deepfake cyberbullying is the use of synthetic media to target a specific person: a classmate, a teacher, a colleague or an ex-partner. The bully takes publicly available photos, videos or voice recordings of the victim and uses AI tools to make them appear to say or do something they never did. The result is shared in group chats, on social platforms or sent directly to the victim with threats.

What makes it different from older forms of bullying is believability and permanence. A fake image or clip looks like evidence. Once shared, copies spread beyond the original group, and the victim may spend months asking platforms to remove them. The FBI's March 2024 public service announcement on AI-generated child sexual abuse material specifically mentions teenagers using AI to create nude images of classmates from ordinary clothed photos.

How common is deepfake bullying among young people?

It is no longer rare. Thorn surveyed 1,200 people aged 13 to 20 for its report Deepfake Nudes and Young People, published on 3 March 2025. One in eight said they personally know someone who has been targeted by deepfake nudes. Only 41% said such images harm the person shown, and of those who saw no harm, 28% reasoned that the images are not real. One in five believed it is legal to create deepfake nudes of someone else, including a minor. Two percent admitted creating them, most having found the tools through app stores, search engines or social media.

In the UK, Ofcom's July 2024 discussion paper Deepfake Defences reported that almost half of respondents aged 8 to 15, and of those aged 16 and over, believed they had seen a deepfake in the previous six months. Ofcom notes that sexual deepfakes overwhelmingly affect women and girls. In the US, NCMEC's generative-AI-related CyberTipline reports rose from 4,700 in 2023 to more than 400,000 in 2025, a figure that includes but is not limited to peer abuse.

What to do if you or your child is targeted

Start by telling the person targeted that it is not their fault, and that they did nothing to deserve it. Then act on the content and the people sharing it, in this order, without making the victim view it again.

If the image is sexual and the person shown is under 18, it is treated as child sexual abuse material in the US even when AI-made. Do not save, screenshot or forward it. Report to NCMEC's CyberTipline and use NCMEC's Take It Down to hash eligible images so participating platforms can block them. Adults depicted in intimate deepfakes can use StopNCII and, in the US, request removal under the TAKE IT DOWN Act's 48-hour process.

• Record where the content appeared: links, usernames, group names, dates and any threatening messages.
• Report the content and accounts on each platform using its impersonation or intimate-image reporting option.
• Tell the school's designated safeguarding lead, or the employer's HR or security team for adult victims.
• Contact police if the content is sexual, involves threats or extortion, or keeps spreading after reports.
• Arrange emotional support: a counsellor, a trusted adult, or a helpline service.

How schools and platforms should handle a deepfake bullying report

Schools need a written procedure before the first case arrives. It should name who receives reports, how evidence is preserved without staff copying sexual images, when police are called, and how the depicted student is protected during the investigation. Discipline for the student who made or shared the content should follow the school's normal due process, with the victim's support needs handled separately and quickly.

Staff can be targets too. In April 2024, Baltimore County Police announced that a recording attributed to the principal of Pikesville High School, which had circulated widely, was not authentic and had been generated with AI. Police said two forensic reviews supported that finding and arrested the school's athletic director. The clip spread widely before that finding, which is why schools should hold off on public conclusions about a viral recording until it has been examined.

Platforms carry obligations of their own. In the US, covered platforms had until 19 May 2026 to set up the TAKE IT DOWN Act's notice-and-removal process for intimate images, including digital forgeries. In the UK, the Online Safety Act gives Ofcom powers to enforce illegal-content duties.

Can deepfake detection settle a bullying case?

Detection can support a case but not settle it. A forensic tool can show that a face was pasted onto another body, that an audio clip carries traces of voice synthesis, or that an image came from a generative model. That helps a school or platform decide how to respond and gives police a starting point. It cannot identify who made the content, and it can be wrong, especially on heavily compressed files that have been re-shared through several apps.

For victims, detection matters less than people think. Removal and reporting routes do not require proof that an image is fake, and a sexual image of a minor must be reported whether it is real or synthetic. Detection is most useful when the authenticity of a clip is itself in dispute, as with a fabricated voice recording of a teacher or a fake video presented as proof of something that never happened.

How NeuroDetect supports investigators and trust and safety teams

NeuroDetect is Neural Defend's forensic workspace for authorized media. An investigator at a school district, platform or police unit uploads the disputed image, audio or video into a case. The analysis records the model and version used and shows detailed views down to regions, frames and audio segments, including explainability heatmaps. An LLM-assisted summary explains the findings in plain language with links back to the evidence, while the analyst makes the final call.

The output is a downloadable forensic report covering observations, uncertainty, provenance, model context and analyst notes. That is the kind of record a fair disciplinary process or a police referral needs, because it states what was found and how confident the analysis is, rather than issuing a bare verdict. Neural Defend's detection runs on ARCAS 1B, its own foundation model, and can be deployed on-premises or air-gapped where student data cannot leave the organization's systems. Sexual images of minors must only be handled under the legal authority that applies to the organization.