OpenAI says child abuse classifier cuts false positives by 48%
🎧 Listen to this story
OpenAI announced a classifier that it says detects child sexual abuse imagery — including AI-generated material — with 48 percent fewer false positives; the images it identifies are reportedly twice as likely to contain abusive material. The company will offer free access to vetted organizations, guidance for developers and training for law enforcement.
The tool is intended to help triage reports at the National Center for Missing and Exploited Children (NCMEC). The agency’s CyberTipline received 400,000 generative AI-related reports in 2025, up from 67,000 in 2024. OpenAI also says ChatGPT will show a warning and impose a 24-hour pause on users who try to generate such material, while reporting all CSAM instances to NCMEC.
Why it matters · editorial interpretation
The tool could help vetted organizations prioritize reports, while the increase in AI-related reports cited in the summary suggests greater demand for triage. Because the figures and capabilities are OpenAI’s claims, its operational impact still needs to be assessed by the institutions that use it.
Sources
SecurityResearch