99 percent. That’s the share of deepfake targets who are women, according to the UN Human Rights Council. Just stop and think about that for a second. Almost every person that’s been targeted by deepfake attack has been a woman.
The number of deefake videos circulating the internet has also grown a lot in the last few years. In 2023 there were around 500,000 instances. Last year that number jumped to 8 million. Volume like that doesn’t stay contained. It spreads. Fast.
For proof, just look at what happened with Grok earlier this year. When Elon Musk’s chatbot added image generation to X, it took users just hours to turn it into an assembly line for creating non-consensual images of women.
Millions of images. Regulators across multiple continents opened investigations. Several countries considered blocking the platform altogether. Lawsuits followed. Months later, despite public promises to fix the problem, journalists were still finding new AI-generated images of real women circulating online.
None of this was unpredictable. It was the result of releasing an incredibly powerful tool without enough guardrails, at a scale no single person could ever hope to fight alone.
So yes, AI has a sexism problem. It’s big, it’s ugly and it’s well-documented.
I came across this issue first-hand a few years ago. It started with a friend. Someone stole her face and body to create fake dating profiles. Those profiles were then used by scammers to con men out of thousands. She wasn’t the criminal, just another victim. Yet she still felt guilty enough that she sent money to one of the men who’d been scammed.
That has always stayed with me. Deepfake abuse steals people’s sense of safety and drags innocent people into crimes they had absolutely nothing to do with. That’s what pushed me to start my company.
I’ve since spent years working in this space, and have come across a ridiculous amount of cases of this happening.
It can be a huge celeb. You might remember the deepfake images of Taylor Swift that spread across social media a couple of years back. But the sad truth is that most victims never make headlines. They’re teachers. Students. Journalists. Nurses. Mothers.
Telegram bots offering deepfake image generation reportedly reached around four million monthly users last year. If that doesn’t tell us this is a systemic problem, I don’t know what will.
So what do we actually do about it? This is where I use a phrase that usually gets a raised eyebrow: ‘AI hair of the dog.’
The idea is simple. The most promising way to fix AI may actually be… AI.
I know how that sounds. The thing that caused the problem is somehow also the solution? But we’ve seen this before. When AI started generating fake online reviews, AI became one of the best ways to detect and remove them. When people worry about AI replacing jobs, many of the most effective retraining programmes also rely on AI to personalise learning.
Again and again, we’ve seen that when technology creates a problem, the same technology becomes the only thing capable of solving that problem quickly enough and at the scale required. Deepfakes are no different.
Detection technology has come a long way. It can now flag manipulated video, audio, and images, and, crucially, explain how they were altered. That matters enormously because if you’re a victim, simply telling people that it’s fake often isn’t enough. You need evidence, something that a social media platform or a police officer can actually rely on.
Humans, meanwhile, aren’t great at spotting convincing fakes on our own. Research has found that when people watch high-quality manipulated videos, detection accuracy can fall to around one in four.
We keep treating deepfakes as an inevitability. Something to brace for, like bad weather. In my opinion, that kind of framing lets everyone off the hook.
AI detection helps to fix this. Not fully or overnight, but meaningfully. If a piece of content can be generated in seconds, it should also be detectable in seconds. If a fake video can spread across a group chat before lunch, there should be a way to remove it before dinner.
Now, I’m not naive enough to think detection tools alone will fix this. We need much better laws to prevent these tools from becoming mainstream. We need more monitoring and defence of dark web sites. Frankly, we need a lot more scrutiny before these legal tools ship in the first place. Verification and provenance standards need to become routine.
But for the first time in years, we also have something we didn’t have before. A genuine countermeasure. AI created many of these problems. It may also be one of our best chances to stop them.
Sometimes the hair of the dog really is the thing that helps the recovery. This time, I hope that’s true.
Julia Jakimenko, Founder & CEO of Cyberette.ai
Julia Jakimenko is co-founder and CEO of Cyberette, an Amsterdam-based company building AI tools to detect and explain manipulated digital media for fraud, forensic, and investigation teams. She holds degrees in Law and Computer Science, and spent 11+ years in compliance, identity, and enterprise risk and security at Rabobank before founding Cyberette. She also built one of the Netherlands’ largest women-in-tech communities from the ground up. She writes and speaks on digital trust, AI governance, and the disproportionate impact of synthetic media on women.
