Deepfakes and Identity Theft
The Emerging Threat to Digital Performers
In January 2023, a Twitch streamer named Brandon Ewing, known to his audience as Atrioc, briefly showed his browser tabs on a live broadcast. One of them was a website selling explicit deepfakes of women he knew personally, including fellow streamers QTCinderella, Sweet Anita, Pokimane, and Maya Higa. The clip spread within hours. QTCinderella, who had considered Ewing a close friend, responded in a tearful stream of her own. "This is what it looks like to feel violated," she said. "This is what it looks like to see yourself naked against your will being spread all over the internet." Her words became a reference point for a problem that has since grown far beyond one website and one friendship. Deepfakes and the identity theft they enable have become an occupational hazard for digital performers, the streamers, cam models, and creators whose faces and voices are their livelihood. Nonconsensual synthetic media is no longer a fringe abuse. It is an industry, and it feeds on the same visibility that performers depend on to earn a living.
The numbers sketch the scale. When the Dutch cybersecurity firm DeepTrace first mapped deepfake videos online in 2019, it found that 96 percent of them were nonconsensual pornography. Four years later, an analysis by Home Security Heroes counted 95,820 deepfake videos online, a 550 percent increase over 2019, and found that 98 percent of all deepfake video was pornographic, with women making up 99 percent of those targeted. Europol's projections suggest around 8 million deepfakes of all kinds were shared in 2025, up from roughly half a million in 2023. Generative tools have collapsed the cost of production to nearly nothing. A few dozen images scraped from a public profile, or a few minutes of stream footage, are enough to build a model of a real person. As journalist Emanuel Maiberg, who has covered this beat since the first Reddit deepfakes appeared in 2017, put it in a talk at Columbia University: once you have the model, you can make images as fast as you can click a mouse.
Performers feel this first and hardest, and their accounts of it are remarkably consistent. Sweet Anita, a British streamer with Tourette syndrome who has never published her real name, described the experience to HuffPost in 2021, before deepfakes of her had even gone mainstream. Strangers had built entire forums around sexualized screenshots and manipulated images of her. "It's basically like having millions of people be your boss, and your boss is allowed to sexually harass you every day," she said. Then the line that should hang over every policy debate on this subject: "I haven't ever even taken my clothes off in front of the camera and yet I'm still a successful porn star." The synthetic version of her had a career she never consented to. She has since become one of the most vocal streamers pushing for stronger legal protections, arguing that no job should carry this level of risk.
There is a second, quieter harm layered on top of the imagery, and it looks more like classic identity theft. In August 2026, USA Today reported on scammers using AI tools to impersonate OnlyFans creators at scale. The scheme is simple. Lift a real creator's photos, generate a synthetic voice, build a fake persona on TikTok, move the conversation to Snapchat, promise a live chat, and collect payment through Cash App, where transfers clear instantly and are nearly impossible to reverse. Then block the fan and vanish. One creator, Jessieanna Campbell, told the paper she constantly fields messages from angry subscribers accusing her of taking money for transactions she never made. Another had fans show up at her home and said she sometimes no longer feels safe leaving it. The fraud has two victims, as the security researchers at Malwarebytes who tracked the scheme pointed out: the fan who loses money, and the performer who loses income, reputation, and sometimes her sense of physical safety.
None of this is incidental to how the creator economy works. It is a direct consequence of it. A streamer's product is presence: face, voice, mannerisms, the feeling of a real relationship with a real person. That presence generates an enormous public archive. A full-time streamer produces hundreds of hours of high-resolution, well-lit, front-facing footage every year, which happens to be exactly the training data a face-swap model needs. The parasocial trust that makes fans willing to pay is the same trust that makes impersonation profitable. Visibility is the job description, and visibility is the vulnerability. Asking performers to protect themselves by being less visible is asking them to earn less, which is why the burden cannot rest on them alone.
Detection Is Losing the Race
The technology built to catch synthetic media is real, and some of it is impressive. Intel's FakeCatcher takes a biological approach, using photoplethysmography to detect the subtle color shifts in facial pixels caused by blood flow, something current generators do not reproduce. It reports 96 percent accuracy in controlled conditions and around 91 percent on deepfakes collected in the wild. Companies like Reality Defender run ensembles of detection models in parallel, scoring content across many signals at once, on the theory that no single model catches every kind of manipulation. But the gap between the lab and the internet is wide and well documented. A 2026 UK government review of the detection market flagged limited representative training data, unreliable accuracy under real-world conditions, and weak incentives for adoption as persistent barriers. Independent researchers have found that tools advertising accuracy in the mid-nineties can perform little better than chance against novel generators, heavy video compression, or content gathered in the wild. Humans are no better. In a 2025 study by the biometric firm iProov, only 0.1 percent of participants correctly identified every deepfake shown to them, even when told in advance to look for fakes. Detection also arrives after the fact. By the time a fake is flagged, it has usually already been seen, saved, and reuploaded somewhere else.
The more practical tool so far has been hashing rather than detection. StopNCII.org, run by the UK charity SWGfL, lets an adult generate a digital fingerprint of an intimate image on their own device; the image never leaves the device, and only the hash is shared with participating platforms, which then block matching uploads. The Revenge Porn Helpline behind it reports a removal rate above 90 percent and more than 300,000 individual images taken down. That model works, but it has two hard limits. It only covers images the victim already possesses, which excludes purely synthetic content fabricated from scratch, and it only covers platforms that choose to participate. Everything else falls to a private takedown industry that charges real money. Female streamers told HuffPost they had paid cybersecurity firms tens of thousands of dollars a year for content removal. After the 2023 scandal, Ewing wired $60,000 to a law firm to fund takedown notices for the women affected and later worked with Ceartas, a service that automates delisting. It is worth pausing on that arrangement: the cleanup of an abuse economy was financed, in effect, by an apology.
Platforms and Lawmakers Respond
Platform policy has moved, though unevenly. Twitch updated its sexual content rules in late 2023 to ban depictions of nudity "regardless of the medium," language aimed squarely at synthetic and doctored imagery, and in 2024 it tightened its sexual harassment policy to explicitly prohibit recording or sharing nonconsensual intimate images, with possible referral to law enforcement. Its AutoMod tool now filters suspected sexual harassment from chat before it appears, leaving the final call to the streamer and their moderators. Meta, TikTok, and others maintain similar prohibitions and participate in hash-sharing programs, and the European Union's AI Act now requires synthetic content to be labeled as such. The gap is enforcement. Reporting regimes still depend on victims finding the content, filing the forms, and waiting, often across dozens of sites, many of them hosted overseas and indifferent to any terms of service. Free expression advocates like the Electronic Frontier Foundation raise a fair counterpoint: automated removal systems are blunt instruments that can sweep up legal content, and 48-hour deadlines leave little room to verify anything. Both things are true, and performers live in the space between them.
The law, after years of lag, has started to arrive. The TAKE IT DOWN Act, signed in May 2025, made it a federal crime in the United States to knowingly publish nonconsensual intimate imagery, including AI-generated "digital forgeries," and required covered platforms to build notice-and-removal systems that act within 48 hours of a valid request. Those platform obligations took full effect in May 2026, with the Federal Trade Commission as enforcer; the FTC has already sent warning letters to 15 major platforms, and the first federal conviction under the Act came in April 2026, an Ohio man who had harassed at least six women with AI-generated deepfakes. The states have moved faster and further. According to Ballotpedia's tracking, 48 states now address sexually explicit deepfakes in some form, up from 46 at the start of 2026. Tennessee's ELVIS Act added voice cloning to the state's right of publicity. Minnesota went furthest, becoming the first state to hold the operators of AI tools themselves liable when their software is used to generate nonconsensual intimate images, a theory now being challenged in federal court.
Two more federal bills would fill the largest remaining holes. The DEFIANCE Act, passed unanimously by the Senate in January 2026, would give victims a civil cause of action with liquidated damages up to $150,000, or $250,000 where the imagery is tied to stalking or harassment, addressing the fact that TAKE IT DOWN criminalizes publication but gives victims no direct way to sue. The NO FAKES Act, advanced by the Senate Judiciary Committee in June 2026, would create a federal right over every person's voice and visual likeness, treating a digital replica as something you own rather than something you chase. Both matter for performers specifically, because their harm is economic as much as personal. Still, every law on the books shares one weakness, and the USA Today investigation stated it plainly: domestic law applies to domestic platforms, and the stolen material largely sits on overseas hosts. Researchers at the University of Bristol added an uncomfortable finding this year, across three experiments, most people kept engaging with deepfake content even after being told it was fake.
Here is where I land after reading through the interviews, the studies, and the statutes. The current model treats this as a content moderation problem, something to be reported, reviewed, and removed one image at a time, and that model has already failed. It places the entire cost of the abuse economy on the people abused: the vigilance, the paperwork, the legal fees, the takedown subscriptions. What performers describe, consistently, is not a series of regrettable posts but a labor condition, a predictable hazard of doing their job in public. Hazards like that are not solved by giving workers better reporting forms. They are solved by making the hazard expensive for whoever profits from it, which in this case means the tool makers, the hosting services, and the sites that monetize the traffic. Minnesota's approach, whatever its fate in court, is the only one that starts from that premise.
Sweet Anita once described herself, bitterly, as a canary in the coal mine, and the metaphor has aged well. What happens to performers online rarely stays with performers. The same tools that fabricate a streamer's body now fabricate executives on video calls, teenagers in school group chats, and job candidates in interviews. Performers simply got there first, because their work made them visible first. They have been saying, in interview after interview, exactly what is coming for everyone else. The record of what it costs is already written. It is written in QTCinderella's exhausted decision to fund lawsuits she should never have needed, in a creator afraid to leave her house because strangers believe she scammed them, and in a woman with millions of followers explaining that she is, against her will, a successful porn star.