By Hamza Faisal

OpenAI has released a new image classifier designed to detect suspected child sexual abuse material, including images generated by artificial intelligence, and said it will offer the tool free of charge to vetted organisations through its API. The company also introduced warnings inside ChatGPT that pause access for 24 hours when someone tries to find such material.

The announcement, made on November 16, is one of the most detailed safety moves the company has published this year. It arrives as AI-generated imagery has made it harder for safety teams to tell real abuse from synthetic material, and as regulators and child-protection groups press platforms to act faster on both.

What the classifier does

The classifier looks for visual signs of abuse in images that have not been catalogued before, including AI-generated ones, and assigns each image a score so that human reviewers can prioritise the most serious cases. OpenAI said it developed and evaluated the tool with the National Center for Missing and Exploited Children (NCMEC), the US clearinghouse that routes abuse reports to law enforcement, and published a model card for it on the same day.

In early testing with partner organisations, the classifier flagged 48 percent fewer images while finding the same number of abusive ones, OpenAI said, meaning reviewers spent less time on material that turned out to be harmless. Images it flagged were twice as likely to be abusive. SafetyKit, one of the early testers, reported that it detected 90 percent of cases involving drawings or animations that sexualise children in its evaluation.

Access is being shared through the Tech Coalition and ROOST, an industry safety body, and vetted child-safety organisations, trust and safety teams and other companies can request it through an intake form. OpenAI also published guidance for developers on November 15, pairing perceptual hash matching, which catches known images, with classifiers for material nobody has catalogued yet.

ChatGPT warnings and 24-hour suspensions

Inside ChatGPT, users who try to find material that sexualises children will now see a warning screen explaining the harm, have the conversation removed, and lose access to the service for 24 hours. The warning also directs them toward specialist support intended to help prevent harmful behaviour. Repeat violations can lead to longer restrictions, permanent account closure, and reports to the authorities.

OpenAI said it shaped the messages with the Finnish child-protection organisation Protect Children, the UK charity the Lucy Faithfull Foundation, and Joel Scanlan of the University of Tasmania, who independently evaluated a trial of the deterrent messages. The interventions are rolling out gradually to personal accounts from mid-November and do not yet cover enterprise or education accounts.

Tools for investigators and developers

Alongside the classifier, OpenAI said it will supply child-safety law enforcement units with AI investigation tools and training, and it has published guidance for developers building on its platform. The company already blocks child sexual abuse material and related requests in ChatGPT, and says it bans accounts and reports attempts to generate or upload such material to NCMEC.

The move is part of a wider push across the industry to give safety teams machine-readable help rather than leaving the work to manual review alone. Whether the classifier meaningfully reduces the volume of abusive material circulating online will depend on how widely other platforms adopt it, but making it free removes one of the main barriers to that happening.