we_are_coded.by CODE · The world, decoded
БГ

OpenAI · Control

OpenAI released a CSAM classifier with AI signal

OpenAI announced on 16 November 2026 an image classifier that detects potential CSAM. According to OpenAI, this is the first CSAM classifier with broad availability that carries a specific signal on whether the material may be AI generated. Access is free for vetted organizations through OpenAI's API.

Source
OpenAI
Published
The event
Checked on
Responsible editor
Cvetelin Ivanov
How we work
MethodCorrections

OpenAI has released a CSAM classifier. According to OpenAI, this is the first CSAM classifier with broad availability that carries a specific signal on whether the material may be AI generated. It is offered free of charge to vetted organizations through OpenAI's API.

Fact

The facts: on 16 November 2026 OpenAI announced a new image classifier that can detect potential CSAM, including AI-generated material. The classifier looks for visual signs of abuse and returns scores that help reviewers decide which images may need attention first. In OpenAI's words, this is the first CSAM classifier with broad availability that has a specific signal on indicating if CSAM may be AI generated. OpenAI's teams trained and evaluated it with assistance from the National Center for Missing and Exploited Children (NCMEC), the nonprofit that operates the CyberTipline. Access is offered to vetted organizations through the OpenAI API at no cost. According to OpenAI, in an evaluation of user-uploaded images the new classifier would flag 48% fewer images while finding roughly the same number of abusive images as the existing tool, and images flagged by it were about twice as likely to contain abuse. According to fal, reviewers confirmed 72% of the classifier's high-confidence flags as novel CSAM. According to invideo AI, one in four flagged cases led to a ban. According to SafetyKit, the classifier detected 90% of cases, with false positives below 1% on legal adult content. OpenAI writes that it recently began participating in the Tech Coalition's Lantern program.

The new part is the separate signal. Until now, detection leaned on the known: a match with material already on file. An AI-generated image matches nothing, because it is new every time. In SafetyKit's words, the classifier catches exactly that kind of unknown CSAM that hash matching misses.

Until now, the tools looked for the known. This one also looks for what no one has seen yet.

The numbers are OpenAI's and its partners'. According to OpenAI, the new classifier would flag 48% fewer images while finding roughly the same number of abusive images. According to fal, reviewers confirmed 72% of high-confidence flags as novel CSAM. According to SafetyKit, the classifier detected 90% of cases, with false positives below 1% on legal adult content. Each of them comes from tests they ran themselves, and the page has no independent check.

Access is requested through an OpenAI form. According to OpenAI, access is given to vetted organizations, but the page does not specify which ones exactly.

The visual is generated code-art, no third-party images.