# OpenAI CSAM Image Classifier — free detection of new and AI-made abuse images

> A free OpenAI classifier helps safety teams find new and AI-generated child abuse images that hash matching misses.

- Org: OpenAI
- Date: 2026-11-16
- Category: security
- Source: https://deploymentsafety.openai.com/csam-image-classifier
- Canonical page: https://ai-tldr.dev/releases/openai-csam-image-classifier/

## In short

OpenAI released a CSAM image classifier, free through its API for vetted organizations, that also flags AI-generated abuse images. ChatGPT now warns users who seek such material and pauses their access for 24 hours.

## What it is

The OpenAI CSAM Image Classifier is a new detection model that scores images for possible child sexual abuse material, including novel and AI-generated images that have never been seen before. OpenAI built it with help from the National Center for Missing and Exploited Children (NCMEC) and offers it at no cost to approved organizations through the OpenAI API. The same launch adds ChatGPT warnings, developer guidance and tools for child-safety police units.

## How it works

Under the hood, an internal CLIP image encoder turns each image into an embedding, and a three-layer MLP maps that embedding to risk scores. The public API returns two scores: likely-csam and likely-csam/gen-ai, which marks suspected AI-generated material. Neither score is a legal finding; the model card says the scores are meant to rank images for trained human review alongside hash matching, not to replace it.

## Why it matters

Hash matching only catches images already known to databases, so freshly generated or new abuse images slip through on image-generation and upload platforms. OpenAI says this is the first widely available CSAM classifier with a separate AI-generated signal, and in its tests it flagged 48% fewer images while finding about the same number of abusive ones, which cuts review work for safety teams.

## Who it's for

trust and safety teams at image and video platforms

## Try it

Request access via the intake form linked from the model card: https://deploymentsafety.openai.com/csam-image-classifier

## Quick facts

- Maker: OpenAI, built with help from NCMEC
- Price: Free for approved organizations
- Access: OpenAI API, after an intake form review
- Scores returned: likely-csam, likely-csam/gen-ai
- Architecture: Internal CLIP image encoder + 3-layer MLP
- ChatGPT change: Warning, flagged chat removed, 24-hour access pause

## FAQ

### Who can use the OpenAI CSAM Image Classifier and does it cost anything?

The OpenAI CSAM Image Classifier is free, but only for vetted organizations. Interested teams request access through an intake form, and OpenAI is working with the Tech Coalition and ROOST to share access. Approved organizations call it through the OpenAI API, and OpenAI says submitted images are processed without retention and are not used for training.

### How well does OpenAI's CSAM classifier perform compared with older tools?

In an OpenAI evaluation on user-uploaded images, the new CSAM classifier flagged 48% fewer images than OpenAI's existing protections while finding about the same number of abusive images, and its flags were about twice as likely to contain abuse. Early partners reported similar gains: fal confirmed 72% of high-confidence flags as novel CSAM, and SafetyKit said it caught 90% of drawn or animated cases.

### What happens in ChatGPT when someone tries to find child abuse material?

ChatGPT now shows a warning that says sexualizing children causes harm, even when the material is AI-generated. OpenAI removes the flagged conversation, pauses the person's ChatGPT access for 24 hours and points them to specialist support. Repeated violations can lead to longer restrictions, account closure or reports to authorities.

### Can the OpenAI classifier prove an image is AI-generated or illegal?

No. OpenAI's model card says the likely-csam and likely-csam/gen-ai scores are risk signals, not legal or factual findings, and the gen-ai score does not verify where an image came from. OpenAI positions the classifier as one layer that works with hash matching, trained human review and legal reporting processes.

## Links

- [Model Card](https://deploymentsafety.openai.com/csam-image-classifier)
- [TechCrunch Report](https://techcrunch.com/2026/11/16/openai-releases-a-new-image-classifier-and-chatgpt-warnings-to-combat-csam/)
- [Developer Guidance](https://developers.openai.com/api/docs/guides/csam-guidance)
- [Techmeme Thread](https://www.techmeme.com/261116/p39#a261116p39)
