OpenAI adds free child-abuse image detection and 24-hour ChatGPT restrictions
Access to the image classifier is limited to vetted organizations. OpenAI estimates it could reduce images sent for review without finding less abuse.
Loading page…
Access to the image classifier is limited to vetted organizations. OpenAI estimates it could reduce images sent for review without finding less abuse.
Listen to this story
OpenAI is adding a prevention and enforcement layer to its child-safety work: a new classifier helps vetted organizations prioritize suspected abuse images, while ChatGPT removes flagged conversations and pauses access for 24 hours after attempts to seek such material. The package also includes developer guidance and AI tools and training for authorized law-enforcement units reviewing records. The classifier is free but access is gated, and its early performance estimates are company-reported; repeated violations can bring longer restrictions, account closure or reports to authorities.
OpenAI estimates the classifier would flag 48% fewer user-uploaded images while finding about as many abusive images as its existing tool; flagged images were about twice as likely to contain abuse.
In partner tests, fal reviewers confirmed 72% of high-confidence flags as novel CSAM, invideo AI banned one in four flagged cases, and SafetyKit detected 90% of cases involving sexualized drawings or animations.
Organizations can use the classifier at no cost, but must be vetted; early results may not generalize to every organization.
Child-safety teams are getting a free tool to prioritize suspected abusive images, while ChatGPT users seeking such material face new warnings and a 24-hour access pause. OpenAI’s November 16, 2026 child-safety announcement combines image detection, account restrictions, developer guidance and investigative support, addressing both the material itself and attempts to seek it.
The new image classifier looks for visual signs of potential child sexual abuse material, or CSAM, including AI-generated images. It can assess previously unseen material and returns scores to help human reviewers decide which images need attention first. The output is a way to prioritize review, rather than a replacement for that review.
OpenAI trained, evaluated and designed the classifier with assistance from the National Center for Missing and Exploited Children, known as NCMEC. The nonprofit operates the CyberTipline, a centralized reporting hub for online child exploitation. Access to the classifier is available through OpenAI’s developer interface at no cost, but only for vetted organizations.
In tests of user-uploaded images, OpenAI estimated its classifier would flag 48% fewer images while finding about as many abusive images as its existing tool. A flagged image was about twice as likely to contain abuse. These early results come from the company and do not guarantee the same performance for every organization.
Partner tests measured different results. At fal, reviewers confirmed 72% of high-confidence flags as novel CSAM in human-reviewed production images, OpenAI says. In invideo AI’s test, one in four flagged cases led to a ban. SafetyKit found the tool detected 90% of cases involving drawings or animations that sexualize children.
OpenAI’s early evaluation estimated fewer images would need review while identifying roughly the same number of abusive images as its existing tool.
ChatGPT already bans CSAM, refuses requests for it and acts against accounts that break its rules. The new warning interrupts attempts to find abusive material, explains the consequences and points users to specialist help. It also addresses AI-generated material, warning that sexualizing children causes harm even when an image is synthetic.
Alongside the warning, OpenAI will remove the flagged conversation and pause ChatGPT access for 24 hours, while keeping specialist support available. Repeated violations may lead to longer restrictions, account closure or reports to authorities. The approach combines an immediate restriction with a route to help people stop harmful behavior.
OpenAI developed the intervention with input from prevention organizations and academic advisers, including Protect Children and the Lucy Faithfull Foundation. Its advisers also included researchers at the University of Kent and the University of Tasmania who independently evaluated a trial of deterrent messages.
The developer guidance covers prevention, detection, review and reporting of suspected abuse. Developed with input from the Tech Coalition, Thorn, NCMEC and the Internet Watch Foundation, it addresses user reporting channels, clear rules and support for people reviewing sensitive content. It also explains how safety identifiers can connect misuse to individual users.
For specialist law-enforcement units, OpenAI is providing AI tools and training. The first tool is intended to review lengthy chat logs and records returned through legal processes, identify potential connections and build timelines. Its findings will link back to source material. Access will depend on agency authorization and program requirements, with program administrators managing access.
OpenAI plans to start investigator training with an overview and demonstration. Teams will also get a field guide to use later. The company says it will propose future training at regional, national and international conferences.
Organizations interested in the classifier can request access through OpenAI’s intake form. The company is working with the Tech Coalition and ROOST to share access, and says feedback from organizations using the tool will inform further improvements.
Loading discussion...
Join the conversation
Explain whether temporary restrictions and specialist support deserve a place in the response.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.