OpenAI launches AI tools to combat child abuse
OpenAI has introduced an image classifier to help detect child sexual abuse material, including AI-generated content, and will offer it at no cost to verified organizations. It has also added warnings and access pauses in ChatGPT, published developer resources and is preparing tools for specialized investigators.

OpenAI has introduced new artificial intelligence tools to detect child sexual abuse material, prevent harmful behavior and help organizations investigating these crimes.
The initiative includes an image classifier available at no cost to verified organizations, new warnings in ChatGPT, a developer guide and support tools for specialized investigators. The goal is for technology to help protect children without leaving important decisions to automated systems.
A detector for finding new material
The new system analyzes visual signals to identify possible child sexual abuse material, including AI-generated content. It also provides a specific signal indicating whether an image may have been artificially created, which can help security teams prioritize new cases.
OpenAI will offer the classifier through its API at no cost to verified organizations. Child protection agencies and companies that need to review large volumes of images will be able to use it, although the system does not replace human assessment.
Early tests show two relevant results:
- It could reduce by 48% the number of images teams need to review, while maintaining approximately the same number of findings of abusive material.
- Images flagged by the new system were nearly twice as likely to contain abuse as those detected with the previous tool.
These figures are initial evaluations, not a guarantee for every service. In tests by partner organizations, fal confirmed that 72% of its high-confidence alerts corresponded to new child sexual abuse material. SafetyKit, meanwhile, detected 90% of the cases evaluated in drawings or animations that sexualized children.
ChatGPT will interrupt attempts to search for abusive content
OpenAI will also add warnings in ChatGPT when it detects attempts to find or sexualize children. The message will explain that this behavior causes harm, even if the requested content is AI-generated, and will provide guidance on seeking help.
The flagged conversation will be deleted and access to ChatGPT will be paused for 24 hours. Repeated violations may lead to longer restrictions, account closure or reports to authorities, depending on the case. The system will also continue to provide specialized support for people who want to stop harmful behavior.
These measures were designed with input from organizations and specialists working to prevent child sexual abuse. They are not intended to turn the chatbot into a therapist or investigator, but to interrupt risky behavior and direct the person toward appropriate help.
Resources for companies and support for investigators
OpenAI also published a developer guide with concrete steps to prevent abuse, let users report problems, review suspicious content and notify the appropriate authorities about possible cases.
The guide explains how to combine different tools, such as moderation systems, specialized detection services and mechanisms for linking multiple incidents to the same account. It also includes recommendations for protecting teams that must review sensitive material.
At the same time, OpenAI will support specialized law enforcement units with training and tools to analyze large amounts of information. The first system will be able to review extensive records of chats and documents obtained through legal processes, identify possible connections and create timelines linked to the original sources.
Access will be limited to authorized agencies and will depend on the program's requirements. The idea is to reduce the time investigators spend organizing data and limit their unnecessary exposure to abusive material.
For you, the most visible change will be in ChatGPT: there will be more blocks and consequences for attempts to search for sexual content involving children. More broadly, the important point is that these tools are also being offered to other organizations, because detecting and reporting this type of material requires coordination between platforms, technology companies, child protection organizations and authorities. The next challenge will be determining whether the systems maintain their accuracy when facing new forms of content and whether human controls continue to have the final say.

