Prosecuting abusive and criminal deepfake images is becoming increasingly difficult: NPR

An Ohio man has been convicted of a cybercrime involving the release of an AI-generated image depicting an abusive sexual act, the first in history under the federal government’s 2025 Take It Down Act. But experts warn that prosecuting such cases is becoming increasingly difficult.

According to the U.S. Attorney’s Office for the Southern District of Ohio, James Straller, 37, pleaded guilty to cyberstalking and publishing digital forgeries for creating obscene visual representations of child sexual abuse, a crime that includes both real and AI-generated images.

The Take it Down Act makes it illegal to publish intimate digital content without consent.

According to the Department of Justice, Straler used dozens of AI platforms and more than 100 AI web-based models on his phone to create more than 700 illegal images for posting on websites dedicated to child sexual abuse material.

Straler was arrested after one of the adult victims reported receiving threatening and harassing messages, according to court records.

Court records say Strahler admitted he was behind the violent phone calls and text messages. Information extracted from confiscated phones revealed additional victims and the scope of his AI abuse.

You can get big dividends with small risk

Kolina Koltai is a senior researcher at Bellingcat, an investigative reporting group, specializing in AI technologies.

He said the sheer volume of content created by Straller is not unusual for this type of criminal and is part of what makes it so difficult for law enforcement to control.

“Even in the early days of AI technology, people are going to have to learn how to install and host things locally on their devices,” Koltai said.

“But today, you can even go to a web domain and type in a prompt. You need very little technical knowledge to start creating content. The overwhelming amount of content makes this a big challenge.”

Koltai cited early editing programs like Photoshop. Photoshop was an early pioneer of amateur image engineering, an expensive graphic design software that required some skill to make realistic edits.

But “now,” she said, “for a dollar or even less, anyone on the Internet can take a photo and put it into Nude Defyer or some kind of AI-generating platform to create a convincing new image, even based on that person’s face.”

Adding to the difficulty for law enforcement when investigating these cybercriminals is the overwhelming number of platforms dedicated to creating deepfake materials.

“It’s often incredibly difficult to know what technology, what services, what platforms a person is using without accessing their device or browser history,” Koltai said.

“It’s not like there are just two or three providers. It’s a multi-million dollar industry, so everyone is trying to get in on the game,” she said, adding that sites often purchase multiple domains under different extensions (.com, .io, etc.) to avoid being taken offline.

“It’s a bit of a hydra situation, even with our research site, where it’s great to shut down a site, but unfortunately there are still a lot of other services trying to take its place,” she said. “It’s a difficult problem to solve until you make it difficult to use these platforms.”

Deepfakes and young people

Matthew Faranda-Diedrich, a lawyer who has handled cases involving deepfake nudes, said AI’s transition from obscurity to mainstream technology has been faster than the law’s ability to adapt.

“Two years ago we’d never heard of anything like this, we’d never seen an incident like this, and now, unfortunately, we’re seeing five or six incidents like this at a time,” he said.

Faranda-Diedrich said he works closely with police to understand rapidly evolving technology and to support the overall investigation of potential illegal activity.

But he said there is often a learning curve for law enforcement and civilians alike to understand how sophisticated many of these apps are at manipulating images inappropriately.

“They’ll think back to when they were younger and other generations and say, ‘Oh, this is like Photoshop,’ and they have this idea in their head that it’s easy to tell that a doctored image is fake. But in reality, the images created by the ‘nudify’ app look very real and look nothing like Photoshopped images.”

“What shall we call it?”

The non-consensual distribution of deepfakes is a multigenerational problem, but research shows it is particularly prevalent among young people.

And women and girls, who make up an estimated 90% of victims of these crimes, are particularly at risk.

Faranda-Diedrich said that in most of the cases he has worked on, both the victims and perpetrators of the crimes were children between the ages of 14 and 16.

“We should try to educate children about this danger and the harm it can cause so that they don’t make ‘stupid’ decisions that actually end up hurting people tragically,” he says.

And schools have a huge responsibility to intervene at the first sign that these technologies are being misused, he said.

“What should we call it? Child pornography,” Faranda-Diedrich said. “And I don’t think a school administrator would ever say, ‘Oh, if only I had known.'” [child sexual abuse material]I won’t call the police. Of course they would. And they need to make the same call here and get it into the hands of law enforcement sooner. ”