OpenAI introduces classifier to identify AI-generated child abuse material
OpenAI has announced a classifier intended to help investigators reviewing child sexual abuse reports identify material that may require action more quickly. The company says the system produces 48 percent fewer false positives than the approach it is replacing. Images flagged by the classifier were twice as likely to contain abusive material, according to OpenAI.
The tool is designed to detect abusive AI-generated images made with different types of software, including open-source models and nudifying apps. John Buckley, OpenAI’s head of child safety public policy, said at a Monday briefing in Washington, DC, that it is the first broadly available classifier for child sexual abuse material to include a specific indicator for AI-generated content.
Buckley said distinguishing images depicting a fictional child from those depicting a real child is important for law-enforcement reporting and triage. He also said AI-generated child sexual abuse material can harm social norms around offending against children and is often found alongside imagery involving real children.
Written by Sakhanda Wire AI from reporting by The Verge. Facts, quotes and full context are in the original article below.