What Is Deepfake Pornography? Scale, Law, and Detection

What Is Deepfake Pornography?

Deepfake pornography is explicit video or imagery created by mapping a real person's face onto another body without consent. The pipeline combines face recognition, face-swapping algorithms, and deep learning models trained to blend the target face into existing footage.

That training step is what separates it from traditional photo editing. A retouched image is a human decision applied frame by frame. A deepfake is a model learning the geometry and lighting of a face well enough to reproduce it under new conditions. The output holds up against casual viewing, and increasingly against the automated moderation systems platforms rely on.

How Big Is the Deepfake Porn Problem in 2026?

The scale is the part most people get wrong.

Prevalence figures come from Sensity AI, which has tracked deepfake video distribution since 2018.

This is not a problem confined to public figures. The entry requirement is one clear photo of a face. Most people have posted one publicly at some point, which means the pool of potential targets is effectively everyone with a social media account. Face-swapped stills follow the same logic, which is why AI-generated image detection matters alongside video.

Why Isn't the Law Stopping This?

Legislation has not kept pace. A handful of U.S. states criminalize the creation and distribution of deepfake pornography. Most countries have no statute that addresses it specifically at all.

The practical effect is impunity. Creators face little risk, so there is no downward pressure on supply — which is why volume keeps climbing rather than levelling off. Enforcement also runs into a jurisdiction problem: the person who made the video, the platform hosting it, and the victim are frequently in three different countries operating under three different legal regimes.

Is AI Making Deepfake Porn Worse?

Yes, in two specific ways. Anthropic's threat intelligence reporting on how AI models were misused through mid-2026 documents both.

Fake identities now go well beyond video

Bad actors are assembling entire synthetic personas: AI-generated profile photos, fabricated biographies, cloned accounts of real people, and messaging convincing enough to sustain live conversations with a victim's genuine contacts. The video is one component in a package built to make the whole thing look verified. Detection that only examines a single file misses the surrounding apparatus, which is the part that lends the video credibility. Our AI scam and deepfake detector is built for exactly this pattern.

Attacks now move in hours, not days

The same reporting describes intrusions progressing from one stolen credential to full system control in roughly three hours. That compression applies to distribution too. Content can be generated, uploaded, and mirrored across multiple hosts faster than most moderation teams complete a single review cycle. Response measured in days has no chance against a cycle measured in hours.

What Actually Stops Deepfake Porn From Spreading?

Scanning at the point of upload. Detection tools that flag synthetic content before it goes live, rather than after it has been shared thousands of times.

Upload is the one chokepoint every piece of media has to pass through. Whatever tool made the file, whoever made it, wherever they are — it has to be uploaded somewhere to reach an audience. That is the single moment where interception changes the outcome instead of documenting it.

Live contexts deserve particular attention. A face swap running in a video call leaves no artifact to analyze later, which is why real-time detection in meetings works on a different principle to file-based scanning.

Detection Alone Isn't Enough

Upload scanning closes the technical gap. Three things still need to move alongside it.

Legislation targeting creation and distribution across more jurisdictions, written to bite before the harm rather than compensate after it.

Platform accountability for moderation systems that are demonstrably failing. Volume is not an excuse when the failure is measurable.

Detection built for full synthetic identities — photos, writing patterns, behavioural signals — not individual videos assessed in isolation. The persona is the attack surface. The video detection layer is one part of it.

FAQ