This is a research prototype. The data and analyses are preliminary and not yet validated — we'd welcome your .
A study by the American Sunlight Project identified over 35,000 instances of AI-generated nonconsensual intimate imagery (NCII) targeting 26 members of Congress, with women being 70 times more likely to be victimized than men. The report highlights the use of generative AI to create deepfake pornography as a tool for harassment and silencing women in public office. The findings underscore the lack of federal legal protections and the disproportionate impact of such abuse on women in leadership.
A study by the American Sunlight Project is reported to have found that 1 in 6 Congresswomen were targeted by AI-generated nonconsensual intimate imagery (NCII) shared on deepfake websites. The study reports having found 35,000 mentions of explicit content involving 26 members of Congress, with 25 being women. Women were 70 times more likely than men to be victimized, according to the report.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
Using AI systems to gain a personal advantage over others such as through cheating, fraud, scams, blackmail or targeted manipulation of beliefs or behavior. Examples include AI-facilitated plagiarism for research or education, impersonating a trusted or fake individual for illegitimate financial benefit, or creating humiliating or sexual imagery.
Also classified as
The targeting disproportionately affected female lawmakers, representing a gender-based misrepresentation and harassment campaign.
The incident involves the targeted creation of humiliating and sexually explicit synthetic imagery of specific female lawmakers to harass and silence them.
Human
Due to a decision or action made by humans
Intentional
Due to an expected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed
The incident was driven by human actors intentionally using deployed generative AI tools to create and distribute nonconsensual explicit imagery targeting female lawmakers.
Added categories peak at 3: Substantial (Psychological Harm, Indirect)
An investigation revealed over 35,000 AI-generated deepfakes targeting 26 members of the US Congress, disproportionately affecting female lawmakers. While not a direct threat to critical infrastructure or state sovereignty, the incident represents a substantial threat to societal stability and human rights, highlighting how generative AI can be weaponized for targeted political harassment and gender-based discrimination.
The AI systems were used as tools by human perpetrators to perform the specific function of generating synthetic images and videos. The reports describe the technology as being used to 'create believable pictures' by 'overlaying headshots onto media of adult performers' or through generative AI, operating on direct human input without autonomous planning or agency.
Limited Risk: The report describes 'AI-generated content such as deepfakes' which are subject to specific transparency obligations under the EU AI Act to ensure users are aware of the synthetic nature of the content.
The AI systems were used to generate nonconsensual intimate imagery and videos of lawmakers by overlaying their faces onto explicit adult content.
The reports are highly consistent, citing the same American Sunlight Project study with identical statistics (35,000+ mentions, 26 lawmakers, 25 women, 1 man). The details regarding the social and political impacts, legislative efforts, and the gendered nature of the harm are clearly documented across multiple reputable sources.
The incident could be viewed purely as a targeted harassment campaign rather than an AI safety failure, but the central role of generative AI in enabling the scale and realism of the abuse makes the AI safety classification primary.
While the specific AI models used are unknown, this detail is not strictly necessary to classify the domain, causal taxonomy, or assess the harms described.
Technology
Generative AI Accessibility
Generative AI makes it easy for perpetrators to create realistic NCII.
Persistent Search Indexing
Google search pages remain indexed even after deepfake content is removed.
Machine Learning Realism
Advanced ML creates highly believable fake pictures, audio, and text.
Data Inputs
Public Availability of Headshots
Headshots of public figures are easily sourced to overlay on adult media.
Target Name Queries
Custom search engines easily locate targets using names and nicknames.
Human Factors
Gender-Based Harassment Motive
Perpetrators seek to shame, strip, and silence women in public life.
Self-Censorship of Victims
Fear of online harassment causes women to self-censor and avoid public roles.
Process and Methods
Ineffective Platform Moderation
Platforms like Instagram fail to act on reported abusive comments.
Unequal Takedown Capabilities
Private citizens lack resources to achieve rapid content removal.
Lack of Control Over Proliferation
Even after takedowns, content continues to proliferate without control.
Management
Failure of Tech Self-Regulation
Critics lack optimism in Big Tech's ability to self-regulate platform harms.
Inadequate Platform Action
Platforms fail to proactively prevent the hosting and sharing of NCII.
Regulatory Environment
Absence of Federal Laws
No federal law establishes criminal or civil penalties for AI-generated NCII.
Patchwork of State Laws
State laws are inconsistent, mostly offering civil instead of criminal paths.
Legislative Gridlock
Bipartisan bills like DEFIANCE and TAKE IT DOWN face hurdles in the House.
