- Research from various cybersecurity firms indicates that the vast majority of deepfakes are pornographic and almost all victims are women.
- One such paper, looking at data from 2023, states that 98% of deepfakes are porn, 99% of victims are women, and that the total number of deepfakes online had increased by 550% since 2019.
- Similar research looking at 2019 data suggested that 96% of all deepfakes are pornography and that 100% of these depict women. A recent major international study about AI and safety also highlighted these results.
- The main caveat around these figures is simply that these issues are relatively new and the technologies used to power deepfakes are increasing in power and proliferation at a rapid pace.
In a 26 January Assembly debate about online safety and violence against women and girls, Alliance Party MLA Connie Egan claimed:
“[Every] day, technologies utilising artificial intelligence facilitate online violence at an increasingly alarming rate. We know that it can happen to anyone — anybody can be a victim — but the evidence points to the fact that it is a gendered issue, particularly when we have data that reveals that 99% of victims of deepfake pornography are women and girls … The ‘2023 State of Deepfakes’ report found that there had been a 550% increase in deepfakes from 2019 to 2023, with 98% of all deepfake videos online being explicitly pornographic. The grotesque use of AI in that way is nothing short of repulsive. Sadly, the creation of deepfakes is not the only use of AI for those who want to harm women and girls online, and we must find ways to keep legislative pace with quickly evolving technological advancements.”
Similar statistics were mentioned by two further MLAs, Diane Forsythe of the DUP and People Before Profit’s Gerry Carroll. Ms Forsythe said:
“A 2023 survey by the online security adviser Security Hero assessed all deepfake videos online, with 98% shown to be pornographic, of which 99% were of women. Its analysis also showed a 550% rise in deepfake pornographic videos since 2019.”
Mr Carroll stated: “The statistics are stark: 96% of deepfakes are non-consensual pornography, and 99% of that material features women.”
There are three key statistics to be checked:
- 98% of all deepfakes are pornographic (Mr Carroll claimed this was 96%).
- 99% feature women.
- There was a 550% rise in deepfakes between 2019 and 2023.
Each of these statistics is supported by evidence.
They are key findings from the 2023 State of Deepfakes report by cybersecurity firm Security Hero.
The discrepancy in the first statistic, whereby Ms Egan and Ms Forsythe both claimed that 98% of all deepfakes are pornographic compared with 96%, as claimed by Mr Carroll, comes from their differing sources.
Ms Egan and Ms Forsythe both cited the 2023 State of Deepfakes, whereas Mr Carroll was referring to a 2019 report (also called the State of Deepfakes). The 98% figure is more recent and – due in large part to this recency, combined with the ongoing proliferation of deepfakes – should perhaps take primacy.
Research in this area appears to be growing. The second annual International AI Safety Report was published yesterday [3rd February] and some information on that, as well as other useful context about deepfakes, is in the article below. Read on to find out more.
- Source
FactCheckNI contacted each of the MLAs about these figures.
Ms Egan pointed us in the direction of the 2023 State of Deepfakes paper cited in her contribution to the debate.
Mr Carroll’s office replied saying that had found such figures via an End Violence Against Women Coalition press release, which cited a report from cybersecurity firm Sensity.
At the time of writing, Ms Forsythe had not responded.
- What are deepfakes?
A paper from the UK government’s Centre for Data Ethics and Innovation (CDEI) published in 2019 states that:
“Deepfakes can be defined as visual and audio content that has been manipulated using advanced software to change how a person, object or environment is presented.”
In previous years this often involved placing one person’s face onto someone else’s body in an existing video (such as pornography) but as the technology behind online image generators has developed this now includes wholly fake videos that are digitally generated.
- Figures
Both Ms Egan and Ms Forsythe cited the same piece of research during their contributions to the Assembly debate. This paper is the 2023 State of Deepfakes by cybersecurity firm Security Hero, which states that:
“In an era of rapid technological progress, deepfakes have become a concern in digital development. This report is a milestone in an ongoing series that thoroughly examines the ins and outs of deepfake technology, revealing its evolving capabilities and imminent threats … Our work stems from a comprehensive analysis of 95,820 deepfake videos, 85 dedicated channels across online platforms, and a meticulous review of over 100 websites linked to the deepfake ecosystem.”
Key findings from the report include that:
- The total number of deepfake videos online in 2023 is 95,820, representing a 550% increase over 2019.
- Deepfake pornography makes up 98% of all deepfake videos online.
- 99% of the individuals targeted in deepfake pornography are women.
- One in every three deepfake tools allow users to create deepfake pornography.
- 48% of surveyed US men have seen deepfake pornography at least once.
- 74% of deepfake pornography users don’t feel guilty about it.
The first three points raised above provide evidence to support each of the three key statistics in this claim.
Other findings include that seven out of the 10 most popular pornography website host deepfakes in some fashion. However, despite this 90% of all deepfake pornography appears on dedicated deepfake websites, with the remaining 10% on mainstream pornography sites.
The report cited by Mr Carroll was The State of Deepfakes, a paper published by Sensity (formerly known as Deeptrace), a deepfake detection technology company based in Amsterdam. This paper was cited by UK group the End Violence Against Women Coalition in a press release critiquing deepfake legislation.
This report was published in September 2019 and states that, at that time, a total of 14,678 deepfakes were online, “an almost 100% increase based on our previous measurement (7,964) taken in December 2018.”
The paper says that, based on Sensity’s analysis, 96% of all online deepfakes are pornographic, while the paper actually suggests that 100% of all pornographic deepfakes analysed at that time featured women.
Figure 1 – source: Sensity
Both these reports use a similar methodology, searching the internet for deepfake forums and deepfake websites as well as, when it comes to pornographic material specifically, analysing some of the most popular porn websites in the world to search for deepfakes.
The report cited by Mr Carroll is from 2019 while the paper noted by Ms Egan and Ms Forsythe says its figures are correct as of 2023.
However, online technology is moving quickly. In particular, the growing capabilities of image generators allows for much easier and faster generation of deepfakes.
This is a rapidly-changing environment, a fact which is illustrated by the significant increase in the total number of deepfakes detected in the report raised by Mr Carroll (14,678) compared with the number detected in the report noted by Ms Egan and Ms Forsythe (95,820) – the 550% increase mentioned in the claim.
- More recent research
Deepfakes are a relatively new phenomenon and, as such, there is not a massive body of research about them.
Both of the reports cited above are papers from cybersecurity organisations rather than strictly academic literature (which does not mean they are wrong).
One significant and relevant piece of research is the International AI Safety Report (IASR), which recently published its second annual study. This work takes a much broader view of online safety – looking at issues ranging from analysis of the increasing capabilities of digital technologies to the risks of these technologies across politics, economics and society – and features input from academics around the world.
The first annual International AI Safety Report was published in 2025, while the second appeared on 3 February this year (after the Assembly debate in this claim took place). Both papers touch on the impact of deepfakes and pornography. The first notes that:
“Abuse using fake pornographic or intimate content overwhelmingly targets women and girls. A 2019 study found that 96% of deepfake videos are pornographic, and that all content on the five most popular websites for pornographic deepfakes targets women (282). The same study found that the vast majority of deepfake abuse (99% on deepfake porn sites and 81% on YouTube) is targeted at female entertainers, followed by female politicians (12% on YouTube).”
The study cited here is in fact the one mentioned by Mr Carroll. The second annual IASR mentions similar statistics.
Other studies cover differents areas, such as research from Ofcom published in 2024 that looked into the public’s exposure to deepfakes.
Overall, and unsurprisingly, the challenge of deepfakes appears to be growing at a significant pace. This article summarises some of the best information that FactCheckNI was able to identify about this issue at the time of writing. If anyone knows of any further research that either exists already or is coming down the pipeline, please get in touch.
