ToolNavs Find Useful AI Tools
Submit Sign in
Back to AI information
OpenAI Opens Its CSAM Detection Tool for Free: One Bad Search in ChatGPT Costs 24 Hours

OpenAI Opens Its CSAM Detection Tool for Free: One Bad Search in ChatGPT Costs 24 Hours

AI information • Admin • • 29 views

On November 17, 2026, OpenAI announced a set of child-safety measures on its website: a free image classifier that can detect CSAM, including AI-generated material, plus new intervention warnings in ChatGPT — users who try to look up child sexual exploitation material are interrupted and lose access for 24 hours.

A free classifier: half as many images flagged, nearly the same catches

Developed with the US National Center for Missing & Exploited Children (NCMEC), the classifier can spot novel images that hash matching misses, and returns scores to help reviewers decide what to look at first. Internal evaluation found it flags 48% fewer images than the existing tool while finding roughly the same number of abusive images, and flagged images were about twice as likely to contain abuse. Early partners reported real-world results: in fal's human-reviewed testing, 72% of high-confidence flags were confirmed as novel CSAM, and SafetyKit saw a 90% detection rate in drawings and animations that sexualize children. The classifier is offered free via API to vetted organizations, regardless of team size.

ChatGPT now interrupts the attempt on the spot

The new warning explains the harm of sexualizing children, including when the material is AI-generated, while the flagged conversation is removed and access paused for 24 hours; repeated violations can mean longer restrictions, account closure, or reports to authorities. OpenAI also published CSAM guidance for developers and is giving law-enforcement units investigative tools and training. The division of labor is clear: detection goes to institutions, intervention lives inside the product, and investigative support goes to law enforcement.

For small safety teams without the staff to build their own tools, getting this level of detection for free is genuinely useful — but the classifier only scores images, and humans still make the final call; access is for vetted organizations, not a public API. For ordinary users, the most visible change is inside ChatGPT: the search is stopped in the moment, not dealt with afterward. Another step on the same safety path is OpenAI inviting outside assessors directly into its labs: OpenAI brings in four external assessment organizations to check model alignment inside its labs.

Recommended Tools

More