The Scale of Non-Consensual Deepfake Imagery: What Recent Data Shows

Deepfake pornography output, victim demographics, and AI-generated CSAM volumes have all risen sharply. Here's what recent research and regulator data actually show.

NE

Noticeora Enforcement Desk · Takedown & Compliance Team

Files DMCA and TAKE IT DOWN Act notices daily across platforms, hosts, and search engines.

Published September 10, 2026 · 8 min read

Non-consensual deepfake imagery — synthetic explicit content generated and distributed without a subject's permission — has moved from a niche technical concern to a documented, fast-growing category tracked by watchdog groups, regulators, and researchers worldwide. The figures below come from platform audits, child-safety hotlines, academic studies, and government data collected between 2023 and 2026.

1. How much deepfake pornography exists, and who it targets

Independent audits of deepfake video hosting sites give a baseline for the scale of the problem and who bears the brunt of it.

MetricValueSource
Deepfake pornography videos identified online (2023)95,820 videos, +550% vs. prior baselineHome Security Heroes
Share of all online deepfake video content that is non-consensual intimate material98%Home Security Heroes
Share of deepfake pornography subjects who are women99%Home Security Heroes
Combined views on top adult sites hosting deepfake content303,640,207 views (7 of top 10 sites)Home Security Heroes
Year-over-year growth in deepfake porn output, 2023 vs. 2022464% increaseSensity AI

98%

of all deepfake video content found online is non-consensual intimate or pornographic material.

Home Security Heroes, State of Deepfakes

Deepfake pornography is not a fringe use case of synthetic media — it is, by volume, the dominant one.

Home Security Heroes, State of Deepfakes report

Distribution channels compound the problem beyond dedicated hosting sites. A Telegram chat room used to request and produce "deepnude" images reached roughly 200,000 subscribers, according to Sensity AI's tracking of dedicated abuse channels.[2]

2. The child-safety dimension

The sharpest recent growth shows up in reports involving minors, where multiple independent trackers converge on the same trend.

MetricValueSource
NCMEC CyberTipline reports involving generative AI, 2024 vs. 20231,325% increase (4,700 to 67,000 reports)NCMEC
Total NCMEC CyberTipline reports, 2024 vs. 2023 (new bundling methodology)20.5M reports, down from 36.2MNCMEC
Same generative-AI report increase, as analyzed by Thorn1,325% year-over-year increaseThorn
IWF-assessed realistic AI-generated CSAM images/videos (2025)8,029 images and videosInternet Watch Foundation
Growth in AI-generated CSAM videos found by IWF, 2024 to 202513 videos to 3,440+ videos (~260x)Internet Watch Foundation
Share of 2025 AI-generated CSAM videos in the most severe legal category65% (2,233 videos)Internet Watch Foundation
Share of victims in IWF's 2025 AI-generated CSAM assessment who are girls97%Rappler

1,325%

year-over-year increase in NCMEC CyberTipline reports involving generative AI technology, 2023 to 2024.

National Center for Missing & Exploited Children

WeProtect Global Alliance's threat assessment describes AI-generated CSAM as continuing to grow "at alarming speed" since it was first flagged as an emerging risk in 2023, a trend it says has been sustained through 2025.[7]

IWF's actionable AI-generated CSAM reports also jumped from 51 in 2023 to 245 in 2024 — a 380% increase[21] — and a 30-day audit of a single dark web forum found 3,512 AI-generated CSAM images and videos, up 17% from a fall 2023 audit of the same forum.[22] For scale context, IWF's total CSAM caseload across all content types — not just AI-generated — exceeded 300,000 reports in 2025.[28] Within that, AI-generated content skews notably more severe: 43% of non-AI CSAM content falls into the most extreme severity category, versus over 65% of AI-generated video content.[30]

AI-generated child sexual abuse material has moved from an emerging risk to a sustained, accelerating threat in the space of two years.

WeProtect Global Alliance, Global Threat Assessment

3. Youth victimization and South Korea's documented crisis

Survey data on teens and case data from South Korea — the country with the most extensively documented deepfake sex-crime caseload — show how concentrated victimization can be among minors.

MetricValueSource
Teens (9-17) reporting some online sexual interaction in 2025, broad definition45% (including deepfake nudes, SG-CSAM, sextortion)Thorn
Overlap between teens affected by deepfake nudes/sextortion and broader interactions67% overlapThorn
Share of South Korea's deepfake sex-crime victims who are children/teens46%MLex
Share of identified South Korean deepfake crime victims who were minors, 2021-2023 police data59.8% of 527 cases (315 victims)Korea Herald
Rise in South Korean government takedown requests for deepfake sex videos, Jan-Jul 2024 vs. 2023~6,500 requests, 4x prior yearNPR
South Korean deepfake porn victimization rate in a 10-country survey3.1%, among the highest surveyedACM CHI

6,500

South Korean government takedown requests for deepfake sex videos in the first seven months of 2024 — four times the same period in 2023.

NPR, citing Korea Communications Standards Commission

Nearly half of the country's identified deepfake sex-crime victims are children or teenagers.

MLex, reporting on South Korea's Personal Information Protection Commission

4. Where non-consensual deepfakes get distributed

Dedicated hosting sites, imageboards, and social platforms all play distinct roles in how this content spreads and continues spreading even after enforcement action.

MetricValueSource
MrDeepFakes peak monthly active users (Nov 2023), before its 2025 shutdown12 million MAUWikipedia, sourced compilation
First major penalty tied to MrDeepFakes distribution, Australia (Sept 2025)A$343,500 fine plus legal feesWikipedia, sourced compilation
Posts on two successor sites after MrDeepFakes' shutdown, vs. its lifetime output359,400 and 104,418 posts vs. 48,734 lifetime postsarXiv preprint
Share of requests on 4chan's dedicated board specifically for non-consensual synthetic imagery86.1%arXiv preprint
Visits to "nudify" sites driven by social referrals over four months (Dec 2025-Mar 2026)5.7M+ visits; ~1.82M from YouTube, 1.3M+ from XInstitute for Strategic Dialogue, via DigiconAsia
Overall image-based sexual abuse reports to UK's Revenge Porn Helpline, 202422,275 reports, up 20.9% from 18,421 in 2023SWGfL / Revenge Porn Helpline

737+

posts per week on one successor site after MrDeepFakes' shutdown — activity that grew, not declined, following the platform takedown and the TAKE IT DOWN Act's passage.

'Deepfake Pornography is Resilient to Regulatory and Platform Shocks,' arXiv preprint

Successor-site posts vs. MrDeepFakes' entire lifetime output

Shutting down the largest dedicated host didn't shrink the ecosystem — it redistributed it, and volume kept climbing.

'Deepfake Pornography is Resilient to Regulatory and Platform Shocks,' arXiv preprint

Even AI chatbots have become a distribution vector: reporting cited alongside IWF's 2025 findings estimated that before safety measures were implemented, the Grok AI chatbot was generating roughly one nonconsensual sexualized image per minute.[29]

Key numbers at a glance

MetricValue
Deepfake porn videos identified online (2023)95,820 (+550%)
Share of deepfake video content that is non-consensual intimate material98%
NCMEC generative-AI CyberTipline report growth, 2023-2024+1,325%
IWF AI-generated CSAM video growth, 2024-202513 to 3,440+ videos
South Korean deepfake victims who are minors (regulator estimate)46%
MrDeepFakes peak monthly active users12 million
Successor-site posts vs. MrDeepFakes' lifetime total359,400 vs. 48,734
UK Revenge Porn Helpline reports (2024), highest on record22,275 (+20.9%)

Sources

Protect your creators before the next leak appears

Facial-recognition detection and legal enforcement, under one flat subscription.

Related reading