Abusing Athletes
4chan Users Target Female Olympians With AI-Generated Non-Consensual Intimate Imagery and ‘Nudified’ Photos
During the 2026 Winter Olympics, users on the anonymous message board 4chan produced and shared AI-generated non-consensual intimate imagery (NCII) of multiple female Olympic athletes, including U.S. Olympians Alysa Liu, Amber Glenn, Isabeau Levito, and Mikaela Shiffrin, as well as Eileen Gu, who competed for China.
For this report, we examined the tactics, techniques, and procedures (TTPs) that 4chan users utilized to create their images. We found that they leveraged low-rank adaptation (LoRA) methods – downloadable files that can train unmoderated, locally run AI models – to generate NCII of the five athletes.
These images perpetuate a form of online harassment that we have previously found was used to produce AI-generated NCII of celebrity and non-celebrity women, including Taylor Swift. The techniques used also reflect how online communities engaged in the creation of AI-generated NCII continue to adapt to the spread of easily accessible AI tools and tighter restrictions on mainstream AI models that limit their use for creating NCII.
While Graphika conducted a separate, independent investigation, the open-source social media intelligence company Open Measures similarly found that 4chan users had generated and “traded” synthetic NCII of female Olympic athletes.
Cristina López G.
Principal Analyst
Cristina López G. was a principal analyst at Graphika, where she led our research into AI-harms and examined social media information operations and networks of online influence. Before her time at Graphika, Cristina managed Data & Society’s Disinformation Action Lab, which focused on networked responses to communications threats specific to the 2020 U.S. Census.
Download the complete PDF
The full report includes the complete network graph maps, raw attribution indicators, cross-platform topology analysis, and the full takedown timeline with platform-level data.
- Full network graph visualizations
- Attribution indicators with confidence scores
- Raw behavioral modeling data
- Takedown coordination timeline
Related Reports

Character Flaws
School Shooters, Anorexia Coaches, and Sexualized Minors: A Look at Harmful Character Chatbots and the Communities That Build Them
Read the Report
Deepfake It Till You Make It
Pro-Chinese Actors Promote AI-Generated Video Footage of Fictitious People in Online Influence Operation
Read the Report
Cheap Tricks
How AI Slop Is Powering Influence Campaigns
Read the ReportSee How Graphika Can Help Your Team Act on This Intelligence
This report is one of 600+ investigations Graphika’s team has published. Our platform gives your analysts continuous access to the same intelligence — plus the tools to apply it to your specific threat environment.