On November 16, OpenAI released the OpenAI CSAM Image Classifier (omni-moderation-csam-latest) through its Moderations API at no cost for approved organizations. The model scores images for signs of potential child sexual abuse material and provides a separate signal for material that may be AI-generated. Safety teams can use the scores to prioritize images for trained human review.
A visual classifier can flag previously unseen images that systems matching files against known material may miss. In an evaluation of user-uploaded images, OpenAI estimated that the classifier would flag 48% fewer images than its existing tool while finding roughly the same number of abusive images. The scores are intended to help reviewers decide where to look first.
Source · openai.com ↗
