OpenAI has released a free image classifier that detects suspected child sexual abuse material (CSAM) — including images that may be AI-generated — and is offering it at no cost through its API to vetted organizations. In ChatGPT itself, people who try to find abusive material now get a warning, lose the flagged conversation, and have their access paused for 24 hours. Why it matters: hash-matching, the industry’s standard defense, can only recognize abuse images that have already been catalogued; a classifier that reads visual signs can flag material that was generated yesterday and has never been seen before.
November 17, 2026
What happened
OpenAI announced the package on November 16. The centerpiece is the OpenAI CSAM Image Classifier, developed and evaluated with assistance from the National Center for Missing & Exploited Children (NCMEC), the US nonprofit that runs the CyberTipline reporting hub. OpenAI says it is the first broadly available CSAM classifier with a specific signal for whether an image may be AI-generated. The release also includes new deterrent warnings inside ChatGPT, published CSAM guidance for developers, and AI investigation tools and training for specialist child-safety law enforcement units.
How the classifier works
The classifier does not issue verdicts. It returns scores — one for suspected CSAM and a separate one for suspected AI-generated CSAM — that help human reviewers decide which images to look at first; a score alone does not confirm abuse or trigger a report. Access is free through the OpenAI API, but only for vetted organizations that apply through an intake form, and OpenAI says images submitted are processed without retention and are not used for training. The Tech Coalition and ROOST are helping share access, and early testers included fal, invideo AI and SafetyKit.
In OpenAI’s own comparison against its existing protections, the new classifier flagged an estimated 48% fewer images while finding roughly the same number of abusive ones — meaning less review work for the same catch — and the images it did flag were about twice as likely to contain abuse. Those are OpenAI’s and its partners’ figures, not independent results.
What changes inside ChatGPT
When ChatGPT detects an attempt to find material that sexualizes children, it now interrupts with a warning explaining that the material is harmful whether AI-generated or not, removes the flagged conversation, and pauses the account’s access for 24 hours, with links to specialist support for stopping the behavior. Repeated violations can bring longer restrictions, permanent account closure, or reports to authorities. OpenAI shaped the messaging with child-protection organizations Protect Children and the Lucy Faithfull Foundation, plus academic researchers who study deterrence. Coverage of the announcement says the interventions began a gradual rollout to personal accounts in mid-November and do not yet cover enterprise or education accounts.
The honest limits
OpenAI’s own model card is candid about the gaps. Precision in an online evaluation drawn from real ChatGPT traffic was substantially lower than in offline tests, recall was weakest for some categories of images, and performance on one service does not guarantee performance on another. The tool also has not been tested on video. In other words, this is a triage aid for human reviewers, not an automated judge — and how much it helps will depend on how widely vetted organizations actually adopt it. The need is not theoretical: NCMEC has said its CyberTipline received more than 400,000 reports involving generative AI in 2025.
Why it matters
For subscribers, the practical takeaway is twofold. If you run any product where users upload or generate images, a free, vetted detection tool from a major lab just lowered the cost of doing this properly — and “we couldn’t afford detection” is becoming a weaker excuse. And if you use ChatGPT, the enforcement is now visible and automatic: a 24-hour pause on a first flagged attempt, with escalation after that. Everything AI, tested — this is the unglamorous side of the technology, and it is the side that decides whether platforms stay usable for everyone else.
Frequently asked questions
Is the OpenAI CSAM classifier really free?
Yes, for vetted organizations. Access runs through the OpenAI API at no cost, but organizations must apply through an intake form and be approved; it is aimed at child-safety groups, trust and safety teams and law enforcement.
What happens if someone seeks abusive material in ChatGPT?
They see a warning, the flagged conversation is removed, and access to ChatGPT is paused for 24 hours, with specialist support offered. Repeat violations can lead to longer restrictions, account closure or reports to authorities.
Does the classifier replace hash matching?
No. Hash matching identifies already-catalogued images; the classifier scores images that may never have been seen before, including AI-generated ones. The two approaches are complementary, and humans still make the final call.
Sources: OpenAI announcement and CSAM Image Classifier model card (November 16, 2026), reported by PPC Land, Digit, Windows Report and Economy Media.

