MIT researchers developed a non-generative evaluation procedure that tests AI models for harmful capabilities without producing illegal outputs, enabling auditors to identify open-source models adapted to produce child sexual abuse material. The National Center for Missing and Exploited Children received more than 1.5 million reports of AI-generated CSAM in 2025, a dramatic increase from 67,000 reports in 2024. The research addresses the challenge that engineers cannot test AI systems for CSAM production capabilities through traditional prompt-and-inspect methods because generating such content is illegal regardless of intent.
Model auditing addresses supply-side risk, but children require protection at the point of actual contact. The research finding—a 2,139% year-on-year surge in AI-CSAM reports—confirms that malicious actors are already deploying fine-tuned models at scale. Guardii closes the operational gap by detecting and blocking AI-generated child sexual abuse material as it travels through direct messaging platforms. Operating across Instagram, Snapchat, Discord and Roblox, the platform's anti-CSAM detection module flags synthetic abuse imagery in real time, surfaces the child to a responsible adult, and enables immediate reporting to the appropriate authority—addressing the harm MIT's research documents at the moment it reaches a target.
