On November 16, 2026 OpenAI announced a suite of tools and resources intended to prevent, detect, and respond to online child sexual exploitation and abuse. The package includes a new image classifier designed to identify potential child sexual abuse material (CSAM), product interventions in ChatGPT, published guidance for developers, and investigator-facing tools and training.
New CSAM classifier and evaluation results
OpenAI said the image classifier looks for visual signs of abuse, including an explicit signal indicating whether CSAM may be AI-generated. The company described it as the first broadly available CSAM classifier with that specific signal. OpenAI reported that the model was trained and evaluated for deployment with input from the National Center for Missing and Exploited Children (NCMEC).
Early tests compared the classifier to existing protections and produced these findings: the classifier could flag 48% fewer images while finding roughly the same number of abusive images; images it flagged were about twice as likely to contain abuse. In partner tests, fal reported that reviewers confirmed 72% of the classifier’s high-confidence flags as novel CSAM, invideo AI said one in four flagged cases led to a ban, and SafetyKit found the classifier detected 90% of drawings or animations that sexualize children while reporting false positives below 1% on legal adult content, according to SafetyKit.
OpenAI will make the classifier available at no cost to vetted organizations through the OpenAI API; access requests are to be submitted via an intake form. The company said it will continue improving the system with partner feedback and cross-platform signal-sharing through programs such as the Tech Coalition’s Lantern program.
Warnings, developer guidance and investigator support
Product changes include ChatGPT warnings that interrupt attempts to generate or find abusive material, explain consequences, and point people to support. OpenAI said flagged conversations will be removed and access paused for 24 hours, with stronger actions for repeated violations up to account closure or reports to authorities. The deterrent design drew on input from organizations including Protect Children, Lucy Faithfull Foundation, Dr. Caoilte Ó Ciardha, and Joel Scanlan.
OpenAI published CSAM guidance for developers, developed with Tech Coalition, Thorn, NCMEC, and the Internet Watch Foundation, and pointed to tools such as the Moderation API, ROOST’s Coop, and Thorn’s Safer. The company also described new investigator tools to analyze chat logs, identify connections, and build timelines; access will require agency authorization and program-level controls, and initial training will be provided with a practical field guide.
Organizations interested in the classifier can request access through OpenAI’s intake form; the offering is restricted to vetted partners and child-safety stakeholders.
Original source: OpenAI News