Forty-seven states have enacted legislation concerning AI-generated child pornography. How did a Wisconsin man’s charge for possessing such materials get dismissed?
Emma Richman, Global Writer
In a federal court ruling late August, judges dismissed a Wisconsin man’s charges of possessing Artificial Intelligence (AI) generated images of children engaging in sexually explicit acts, citing First Amendment protections.
Steven Anderegg, a software engineer from Wisconsin, used an open-source AI image generator to create more than 13,000 photorealistic images depicting child sexual abuse, also known as Child Sexual Abuse Material (CSAM).
According to child rights organization Enough Abuse, similar Supreme Court rulings of the past forced the federal court to decide on a difficult conclusion: “Material involving an actual child (deepfakes) may be prohibited even in the privacy of the home, while wholly synthetic imagery is entitled to substantially greater constitutional protection.”
While this narrow ruling doesn’t legalize the possession of AI-generated CSAM, it raised alarm bells from child rights organizations and social media users on what this means for similar cases, as AI technology continues to grow exponentially.
Christy Delafield, Senior Director of Communications for the Washington, D.C.-based nonprofit Together For Girls, shares this sentiment.
“One in eight children globally have been subjected to online solicitation and have experienced non-consentual taking, sharing and/or exposure to sexual images and videos,” Delafield said in a statement to Miss Reporter News, citing a study from Childlight. “AI uses these images; AI generated child sexual material is not a victimless crime.”
The Larger Issue, and a Myriad of Difficulties
Facebook family photos are shared by family members; images of birthday parties, graduations and sports competitions get posted on Instagram by proud parents.
AI image generators train themselves by scraping the internet for publicly accessible photos, illustrations and text — like the ones listed above — and use that material to generate images in response to a given prompt. That all means one thing: Everything generated by popular tools like ChatGPT and Stable Diffusion, including AI-generated CSAM that doesn’t depict any one child, is based on an amalgamation of real media found from across the web.
Though it is debated whether or not image generators can produce entirely original images, studies by child rights organizations show that the bridge between artificial and real CSAM continues to shrink.
In their most recent annual report on AI-generated CSAM, the Internet Watch Foundation (IWF) uncovered real photos of child sexual abuse within datasets used to train AI models. Users can fine-tune these tools, which leads the IWF to conclude that AI-generated CSAM causes real harm. As the organization puts it, tech-facilitated CSAM “fuels sexual interest in children, normalises extreme violence, and increases the risk of contact offending.”
Stephanie Hall is the Technology and Innovation Lead at the Minnesota-based nonprofit Zero Abuse Project, where she trains industry professionals on online child safety and emerging technology. For Hall, this includes fighting against the assumption that “synthetic” automatically means “harmless” when it comes to AI-CSAM.
Hall noted in an interview with Miss Reporter News that when investigators comb through platforms where users spread AI-generated CSAM, they often find real CSAM — blurring the line between perpetrators who view the synthetic version and those who view the real thing.
“Even if there isn’t an identifiable child depicted in one particular image, that material still exists inside a much larger ecosystem that involves real offenders and real victims, real CSAM,” Hall said.
She went on to say the question shouldn’t stop at whether or not a real child was depicted in a particular image.
“I think we also need to ask how the material was created, how it’s being used, where it’s being exchanged, and what role it plays in the broader exploitation ecosystem,” Hall argued.
Back to the Case
The Justice Department was notified of Anderegg’s crime in 2024 when he sent AI-generated CSAM to a minor via Instagram DM. Under a 2024 Wisconsin ban on AI generated CSAM that extends to material not depicting real children, the Justice Department charged Anderegg with creating, possessing and distributing such materials.
In August the case was brought to the U.S. Court of Appeals, where federal Judge John Lee ruled the Wisconsin law unconstitutional when applied to Anderegg, dismissing the possession charges. Lee argued the Supreme Court has determined that the First Amendment enables individuals to possess obscene materials in their own home. He also cited a similar 2002 ruling in his decision, which extended that protection to virtual child abuse material that doesn’t depict an actual child.
Enough Abuse reported that 47 states have enacted laws criminalizing or banning AI-generated or computer-edited CSAM. The Trump administration’s TAKE IT DOWN ACT, which criminalizes sexual deepfakes, or AI-generated images depicting a real person, also became law in 2025.
University of Wisconsin Law School Professor Ben Sobel, who examines the intersection of law and technology, believes such laws aren’t always clear in what they seek to regulate. Some focus on materials considered “obscene,” while others seem to require that the material deceives people into thinking it’s recording factual information.
“I just think that there are crucial ambiguities in the way that these statutes are written, and it suggests a larger uncertainty about what they are meant to do that needs to be clarified,” Sobel said in an interview with Miss Reporter News.
Sobel also acknowledged the unique complexities in cases like Anderegg’s, where the crime doesn’t involve an identifiable victim. When the crime involves a deepfake of an identifiable person, one can make the argument in court that the material caused emotional distress or deception of the public. When the material depicts individuals that are AI-generated, whether children or otherwise, Sobel remarks that this is not the case.
“I think it would be helpful to revisit the literature on the harms of pornography generally, and feminist scholarship on those harms,” Sobel said on how legislation could address AI generated pornography without identifiable subjects. “Because the theorists can posit that the existence of pornography harms women as a class, for example, and so one could certainly make that kind of argument about pornographic AI generated depictions, even if they don’t depict a particular person or woman.”
Sobel believes this is why there hasn’t been laws targeting AI-generated sexual images of any kind, but rather deepfakes of identifiable people or CSAM, and that considering these arguments may bring legislatures closer to making such laws.
Anderegg still faces charges for the creation and distribution of AI-generated CSAM, but the application of the First Amendment in the ruling offers another example of the difficulties that surround criminalizing AI image generation — such as free speech, lack of regulation, conflicting legislation and rapidly growing technologies.
The federal judges noted a concern for the latter within their decision, admitting that evolving technologies complicate the lines drawn by precedents set by Supreme Court cases, the most recent being in 2002.
Part of the court ruling reads, “We now live in an age where GenAI models can render images depicting the abuse of virtual children that are virtually indistinguishable from those depicting the abuse of actual children.”
While Sobel believes some AI legislation is unclear in what it seeks to regulate, Hall posits a possible clarification: Legislatures should not treat AI-CSAM as an AI policy issue, but a “child safety issue, a criminal justice issue, a platform safety issue and a survivor issue.”
Fighting Tech-Fueled Fire With Fire
Worldwide child safety organizations have had to adapt their efforts to this rapid growth in AI technology and have expanded their case load concerning child abuse and exploitation to include reports of deepfakes and AI-generated CSAM.
This starts at the initial investigation. Hall works closely with investigators from the national Internet Crimes Against Children Task Force, where they combed through more than 400 thousand CyberTipline reports involving generative AI last year, Hall noted in a LinkedIn blog post. Usually, their investigations begin with just an IP address of a perpetrator, then they must determine whether the abuse material is real or synthetic. This is while battling strained time and dwindling resources according to Hall, depending on the amount of investigators stationed in a city’s law enforcement agency.
This is one of the reasons behind NOVA, an AI tool for investigators that Hall’s team created. Its purpose is to help streamline the initial investigation process by using a perpetrator’s IP address or username to find much more information, like other social media accounts, an address, telephone number and relatives.
In her work exploring the ways child safety and technology can intersect, and teaching the harms of that technology, Hall emphasizes a “human-centered approach” to AI use.
“My goal in all of my trainings and speeches is to talk about how two things can exist at once,” Hall said. “AI can be incredibly helpful to non profits like us, just allowing us to do things that we couldn’t typically do and allow us to reach more people and do more good. At the same time, it can also cause an extreme amount of harm, specifically in child abuse prevention.”
In July, the global nonprofit Safe Online announced it will give $8 million in grants to 30 nonprofit organizations in the latest rollout of funding to end online child sexual abuse and exploitation. Many of these organizations focus on creating similar AI-assisted technology to respond to “emerging technology risks.”
Among the recipients, Kindred Tech and Aselo’s AI platforms are similarly aimed at helping caseworkers and law enforcement investigate crimes from CyberTips, while the nonprofit Thorn is using AI to accelerate victim identification.
Delafield works at Together For Girls, which is using its Safe Online grant to put together a global Safe Tech Lived Experience Council. The ten-member advisory group will be made up of survivors of online sexual abuse, whose goals are to influence global policy and ensure child safety by working with technology companies, governments and regulators. Launching spring of 2027, Delafield says it will be the first global advisory body of its kind.
In an interview with Miss Reporter News, Delafield emphasized how technology companies — like OpenAI, Meta and Anthropic, among others — control many important avenues of improving online safety. “That’s why our Safe Tech Lived Experience Council was started,” Delafield said. “So that tech companies and those that regulate them can hear directly from people who have experienced harm, and develop safety by design.”
Lee’s ruling cemented what Delafield knew already: The law is not keeping up with technological advancements. In a statement to Miss Reporter News, she argued that the corporations behind the biggest advancements in AI need to work with global policy makers to enact these child safety-first regulations.
“Any emerging technology carries risks, but those risks can and should be mitigated. Both governments and the tech industry have a role to play in ensuring safety by design,” Delafield said. “The question is: Will we meet [at] this moment?”
Leave a Reply