SAN FRANCISCO - OpenAI released a new image classifier on Nov. 16 to detect child sexual abuse material, including AI-generated imagery, alongside 24-hour ChatGPT account pauses and free API access for vetted safety organizations.
Free Classifier Targets Novel AI-Generated Abuse
Developed with the National Center for Missing and Exploited Children, the tool scores visual signs of abuse in previously unseen images and flags whether material may be AI-generated. OpenAI is offering free API access through partners Tech Coalition and ROOST. In internal evaluations, the classifier flagged 48% fewer images while catching roughly the same volume of abuse, doubling the likelihood that flagged items contain harmful content. Partner SafetyKit reported 90% detection on abusive drawings and animations with under 1% false positives on legal adult content.
ChatGPT Adds 24-Hour Suspensions and Support Links
OpenAI is rolling out in-app warning screens to personal ChatGPT accounts globally to interrupt prompts seeking sexualized content involving minors. When triggered, ChatGPT deletes the flagged conversation, suspends access for 24 hours, and directs users to specialist intervention helplines, with repeat violations risking permanent bans or law enforcement referrals. Researchers at the University of Tasmania and the Lucy Faithfull Foundation helped design and evaluate the deterrent messages.
Developer Guidelines and Law Enforcement Tools
The company also published developer safety guidelines and joined Anthropic and Google in commissioning Tech Coalition guidance on handling abuse disclosures. For child-safety investigators, OpenAI is introducing authorized plugins and data analysis tools to organize chat logs and legal returns.


