MediaNama reported that OpenAI has released a classifier that flags child sexual abuse material (CSAM) in images, including AI-made or edited ones. OpenAI built it with the US National Center for Missing and Exploited Children (NCMEC). Approved groups can use it free, including platform safety teams, non-profits and law enforcement.
It gives two risk scores: suspected CSAM and suspected AI-made CSAM. Scores help reviewers choose what to check first. They are not verdicts. In offline tests, the suspected-CSAM score had 99.1% precision and 80.2% recall, so roughly one in five abusive images in that test went unflagged. On ChatGPT traffic, its precision was 29.8%, so roughly seven in ten flagged images were not CSAM. OpenAI attributes part of the gap to how rare abusive images are in real traffic.
Precision depends on how common abuse is in a platform's images. When it is rare, false alarms make up a bigger share of the flags. So a vendor's lab figure says little about your own review queue. OpenAI advises organisations to test the tool on their own data first. It also says human review stays necessary for reporting decisions.