Artificial intelligence (AI) has significantly lowered the barrier to entry for the production and distribution of nonconsensual intimate images (NCII) and child sexual abuse materials (CSAM), leading to a network of abusive tools, malicious actors, and supporting infrastructure at scale that is extremely difficult to dismantle through existing regulatory approaches.
A new CLTC white paper proposes recommendations for policies that can be implemented to regulate technologies that enable the creation and dissemination of AI-powered NCII and CSAM. The report maps existing laws against the AI NCII and CSAM value chain (i.e., the individuals and platforms supporting the production and distribution of these materials) and identifies regulatory gaps that lawmakers can address.
The report, Closing Gaps Across the Ecosystem: Regulating the Technologies that Enable AI-Powered Nonconsensual Intimate Images and Child Sexual Abuse Materials, was authored by Charlotte Yuan, Nonresident Fellow with the AI Security Initiative (AISI), and Nada Madkour, Director of AISI. The authors conducted the research as part of the program of professional education at UC Berkeley’s Goldman School of Public Policy.
“Robust policy guardrails are needed to mitigate the harms of AI-powered NCII and CSAM, which threaten individuals’ privacy, financial stability, reputation, and psychological well-being,” the authors explain in the report’s introduction. “However, the regulatory landscape in the U.S. remains fragmented and is insufficient to address these risks. Despite a recent surge in legislative developments around deepfakes, lawmakers continue to struggle with determining and prioritizing the most critical intervention points.”
While U.S. lawmakers have made efforts to prevent NCII and CSAM — including through the TAKE IT DOWN Act (TIDA) a federal law was signed into law in May 2025, criminalizing the nonconsensual sharing of authentic or AI-manipulated intimate images — this law has proved to be insufficient in addressing the challenge. “Most laws focus on prosecuting individual creators and distributors, while the ecosystem of technologies, platforms, and economic incentives that enable the mass proliferation of AI-powered NCII and CSAM remains largely unregulated,” the authors write.
Peeling Back the Layers
To produce the paper, Yuan and Madkour conducted a review of prior research, engaged with experts, and analyzed existing legislation, including 28 federal and state laws in California, New York, Texas, and Utah. Their paper identifies where current regulations fall short in mitigating harms, and which nodes in the value chain have the greatest potential for regulatory intervention. “Current legislative efforts overwhelmingly focus on punishing individual creators and distributors, which limits accountability for key actors such as social media companies, open-source platforms, and model developers,” they explain.
The report draws on a three-layer framework introduced by Bailey and Barrington comprising “Model, Deployment, and Distribution” components of AI regulation (Bailey and Barrington, 2026: under submission), and provides an overview of the larger ecosystem of technologies that facilitate the proliferation of abusive content. In particular, the analysis adapts each layer to the CSAM/NCII regulatory context, and adds an additional “human” layer to incorporate creators and distributors.
- The human layer includes individual creators and distributors of AI NCII and CSAM.
- The model layer includes the technologies that support development of AI models and apps, including training datasets that contain harmful content, open model releases, and open-source platforms that host harmful training datasets.
- The product deployment layer represents the software environment and user interface through which a model or app is accessed by end users (e.g., a web browser or app), including generative AI or text-to-image/video models, payment processors, infrastructure providers, app stores, and search engines.
- The distribution layer includes all technologies that enable the distribution or circulation of AI NCII and CSAM content, including social media platforms as well as public and private platforms.
The recommendations proposed in the paper prioritize the deployment layer, which has the “most opportunity for intervention,” followed by the distribution layer, and lastly the model layer. “A major contribution of the paper is to propose policy solutions that proactively prevent the generation of abusive materials from the beginning of the AI NCII and CSAM lifecycle, rather than simply containing or removing the distribution of such content on the internet,” the authors write. “By mapping existing regulations to potential intervention points in the deepfake ecosystem, we aim to help policymakers identify and address hidden technological nodes and actors that remain underregulated.”
A figure showing the percentage of state and federal legislation corresponding to different layers in the AI NCII and AI CSAM value chain, adapted from Bailey & Barrington’s three-layer AI regulatory framework (Bailey and Barrington, 2026: under submission).
Recommendations for Regulation
The paper offers recommendations for proactive regulations at both the federal and state levels that would target the structural conditions that allow AI CSAM and NCII to proliferate.
Federal-Level Regulation
Remove the requirement that AI CSAM content depict real children for it to be illegal. The requirement of proving that AI-manipulated depictions are indistinguishable from real children creates a significant loophole for CSAM creation and distribution, as it allows content to be proliferated as long as it is “not realistic enough.” To strengthen existing protections, future legislation should remove such a requirement.
Impose liability on platforms that host harmful models or training datasets. Many open-source platforms host models and training datasets that contain CSAM and NCII materials, allowing malicious actors to build models designed for nudification. Platforms should be required to implement a mechanism for the public to submit reports and establish response procedures, including processes for reviewing submissions and promptly taking down models or training datasets that contain or facilitate the creation of harmful content.
Eliminate the requirement that perpetrators must have intent to distribute. A major gap of the TAKE IT DOWN Act is that it does not address perpetrators who create content for personal use, but rather focuses on those who distribute materials on public channels, and it requires that intent be established at the time when the content was produced. Future federal statutes should eliminate intent to distribute as a prerequisite for liability.
Establish a private right of action to allow victims to sue perpetrators. TIDA consists of both criminal and civil components, as it criminalizes AI NCII and CSAM while requiring covered platforms to establish notice-and-removal mechanisms under the enforcement of the FTC. However, the law does not include a private right of action allowing a victim to file a lawsuit to seek compensation for harm. Criminal cases often present high barriers for victims navigating the court system and typically do not provide compensation for damages. Therefore, a victim-centered civil law that guarantees the right of action is essential to fill these gaps and lower the barrier to entry for smaller cases.
State-Level Regulation
Ban advertisements for AI nudify apps on social media platforms. Social media platforms make millions of dollars each year promoting advertisements for “nudify” apps that provide tools to create sexually explicit materials without consent. An outright ban on hosting advertisements for such services would effectively reduce users and thus the proliferation of abusive content. With faster legislative cycles, states can pass legislation to limit nudification apps.
Extend liability for AI NCII and CSAM to launch infrastructure providers. Launch infrastructure that supports the commercialization of apps or models includes critical infrastructure providers, such as cloud platforms and app stores, which facilitate the monetization and proliferation of deepfake models and apps. Only 4% of enacted laws in the four states analyzed provide pathways for prosecuting infrastructure providers that host harmful content. To fill this gap, states should design interventions that require each actor in the launch infrastructure to take appropriate mitigation actions.
Require companies to deplatform harmful materials within a specific time frame. AI has amplified existing risks and harms of NCII and CSAM by increasing access to tools used to generate materials, lowering technical barriers, and making synthetic content more realistic. To control the proliferation of these materials, states that do not currently have take-down laws should establish mandatory removal of AI NCII and CSAM from platforms that host the content, especially social media platforms and deepfake sites, and states should also mandate that platforms take down content that facilitates malicious actors from practicing sextortion.
Enhance data integrity and content provenance across the model lifecycle. Policymakers should enforce data integrity and content provenance obligations across both input and output sides of model development, requiring developers to verify that training datasets are free of illegal material and to document the origins of generated content. On the input side, states should mandate that all AI training data undergo standardized, pre-training cryptographic hashing, a process that turns images and videos into unique digital fingerprints. On the output side, policy should mandate that AI developers comply with content provenance standards that allow tracing AI-generated content back to its source.
Promote AI literacy to combat harms in schools. Traditionally, CSAM and NCII have been created by adults, but with the increased accessibility of digital tools and reduced barriers to entry, the younger generation can now use AI tools to create content depicting their peers and adults. The increasing generation of harmful content by minors complicates the regulatory landscape and calls for improved education on deepfake technologies. This may include teaching and training students about the risks and harms associated with deepfake technologies, the importance of consent, and the disproportionate impact on women and children.
“AI has transformed the creation and distribution of NCII and CSAM from a problem of individual bad actors into an industrialized ecosystem that spans multiple technologies and platforms. Most current laws have not been updated to reflect this new reality, however,” the authors write in the conclusion. “This analysis mapped 28 enacted laws onto the NCII and CSAM value chain, revealing that enforcement is concentrated on creators and distributors, with a limited number of regulatory interventions targeting the platforms and infrastructure that enable the mass production, distribution, and monetization of content. The recommendations in this paper aim to help shift policy toward a more effective framework that spans the entire AI NCII and CSAM value chain.”
