OpenAI opens its CSAM image classifier to vetted groups for free, with a dedicated signal for AI-generated abuse
Synopsis
OpenAI released an image classifier that detects child sexual abuse material (CSAM) and says it is the first broadly available CSAM classifier with a specific signal for whether an image may be AI-generated. In an evaluation on user-uploaded images, the new classifier was estimated to flag 48% fewer images than existing protections while finding roughly the same number of abusive images, and flagged images were about twice as likely to contain abuse. Access is free through the OpenAI API to vetted organizations, and the National Center for Missing and Exploited Children helped with training and evaluation.
Interpretation
A CSAM image classifier for child-safety teams and platform reviewers is now broadly available, carrying a dedicated signal for whether an image may be AI-generated. Existing protections already block and report abuse, but the new tool targets images not seen before and helps reviewers prioritize those more likely to contain abuse. In an evaluation of user-uploaded images against existing protections, the new classifier was estimated to flag 48% fewer images while finding roughly the same number of abusive images, and flagged images were about twice as likely to contain abuse.
The classifier looks for visual signs of abuse and returns scores that let reviewers order images, rather than issuing a final judgment. The scores decide which images need attention first, concentrating human review on a smaller, more suspicious set. Early partner tests covered different content: fal confirmed 72% of the classifier's high-confidence flags on human-reviewed production images as novel CSAM; one in four flagged cases in invideo AI's testing led to a ban; SafetyKit detected 90% of cases involving drawings or animations that sexualize children.
Perspective
This report draws on OpenAI's release note and covers the classifier's availability, its evaluation figures, the ChatGPT warning intervention, CSAM guidance for developers, and tools and training planned for child-safety law enforcement units. The classifier is offered through the API only to vetted organizations and requires an access request; use of the law enforcement tools depends on agency authorization and program requirements.
The release says early tests suggest the classifier can help reviewers focus on abusive images while reducing review of false positives, wording that keeps the result preliminary. The classifier will continue to be improved with user feedback, and future iterations of the law enforcement training are to be proposed at conferences without described content.
