We are committed to sharing our data and our codebooks with the Artifact Evaluation committee to ensure our artifacts meet the USENIX Open Science standards. This suggests that laws are powerful enough to disrupt behaviors of even highly skilled, core members of the community.We argue that augmenting our existing mitigation approach by incorporating legislative protections has significant potential to more effectively stem deepfake abuse. The timing of the shutdown seems linked to the new Take It Down Act that Congress passed last week. This law criminalizes "non-consensual intimate imagery," including AI-generated content, and requires websites to remove such content within 48 hours when someone complains. We are committed to promoting tools and resources that align with ethical standards and respect for privacy. Our recommendations focus on products and services intended for legitimate, creative, and professional purposes.
It is not just Bollywood actresses who are affected - others have been targeted recently, including the Indian industrialist Ratan Tata, who had a deepfake video made of him, giving investment advice. They are the latest in a string of deepfake videos which have gone viral in recent weeks. We are committed to sharing our data and code with other researchers at academic, non-commercial institutions seeking to conduct future work.
To mitigate these harms, best free porn videos the researchers had regular check-ins regarding their psychological well-being, worked together as a team to process MrDeepFakes content, had access to therapy and counseling services, and took breaks as needed. For the community favorite, DFL (the Github repository was linked 203 times), the corresponding paper was initially published as an arXiv preprint in 2020 [49] and later published in the peer-reviewed journal Pattern Recognition in 2023. The delta between the two versions largely focuses on potential abuses of DFL, which both versions claim is most commonly used by VFX artists. Neither version directly addresses the use or abuse of deepfake technology to generate sexual imagery. In August 2024, the repository for DFL was disabled by GitHub for terms of service (ToS) violations. According to the site footer, MrDeepFakes’ stated purpose is to "provide a safe-haven without censorship" for people to experiment with deepfake technology, within the rules of the community.

What are some ethical concerns related to deepfakes?

His Tom Cruise deepfakes, which overlay the young actor’s face on a lookalike with a similar build and haircut, don’t attempt to fake Mr Cruise’s ears. Deepfakes made in any of these ways can be convincing, especially if they’re tidied up using conventional editing software afterwards. Since they can now be made so easily, demanding relatively little time, effort or technical knowledge, it’s more likely than ever that you might encounter one that looks genuine. Did Labour leader Sir Keir Starmer yell profanities at his staff, or endorse an investment scheme? Did Mayor of London Sadiq Khan say that "Remembrance weekend" should be postponed in favour of a pro-Palestinian march?

Likewise, one buyer clearly expresses violent attitudes toward women, stating that the creator would get "bonus points" if they can make it "rough" or "painful," tying the buyer’s sexual gratification to violence against women. Prior work has predominantly described the motives for deepfake abuse as either seeking nonconsensual sexual gratification or producing disinformation [9, 61, 45]. While we do find sexual gratification to be a major motivator, we find others as well. Since then the technology - which relies on complex machine learning algorithms - has evolved rapidly and deepfakes have become more common.

While the names of his victims are supposed to be kept confidential, he also forwarded the court order to 49 email addresses, including to media outlets, which identified some of the victims. A deepfake video of me, in the BBC News studio, apparently presenting a "breaking business news" story has been swirling around social media. The world's biggest YouTuber, MrBeast, and two BBC presenters have been used in deepfake videos to scam unsuspecting people online. In examining user data, we collected only publicly available data, and the only potentially personally identifying information we collected was the account username and the user ID. We never attempted to deanonymize any user in our dataset and we did not interact with any community members in any fashion (e.g., via direct messages or public posts). In our report, we take careful note to not include any users’ account details when reporting quotes and opt to only report aggregate statistics to protect user privacy.
The security community has taxonomized the harm of online abuse, characterizing perpetrators as driven by the desire to inflict physical, psychological, or sexual harm, silence, or coerce targets [56]. In this paper, we systematically characterize the MrDeepFakes platform over the past five years. We document the rise of this online community and analyze the economics of the marketplace, focusing on the buyers, sellers, and targets of deepfakes posted to the site. Through a detailed analysis of forum discussions, we also examine the state-of-the-art technical components required to create deepfakes.
The user Paperbags — formerly DPFKS  — posted that they had "already made 2 of her. I am moving onto other requests." "In a really serious way. It really discourages people from going into politics, going, even being a celebrity." In 2025, she said the technology has progressed to where "someone who's highly skilled can make an almost indiscernible sexual deepfake of another person." "A critical service provider has terminated service permanently. Data loss has made it impossible to continue operation," says a shutdown notice on the website's main page.

"We are already at the point where you can't tell the difference between deepfakes and the real thing," Professor Hao Li of the University of Southern California tells the BBC. If you do come across any suspected deepfakes online, please do send them our way for fact checking. As AI technology advances, the above tips for spotting deepfakes will likely date fast. Like most of these tips, the presence or absence of this isn’t a guarantee either way—the deepfake might not include this error, or the real person may have a squint. Mr Ume pointed out that there are often issues with the alignment of people’s eyes in deepfakes. He also commented that it’s difficult to fake ears, as they’re so specific to the person.