Skip to content
Advertisement
OpenAI rolls out image classifier and ChatGPT warnings to fight CSAM

OpenAI rolls out image classifier and ChatGPT warnings to fight CSAM

OpenAI official logo (public domain, Wikimedia Commons) — CryptoBriefing brand treatment

The company is pairing detection tech, in-chat warnings, and developer guidance as AI-generated abuse material strains older safety tools

OpenAI has announced a new set of tools aimed at child sexual abuse material (CSAM). The centerpiece is an image classifier built to detect such content, including images that were generated by AI.

The rollout has four parts: the new image classifier, warnings and temporary blocks inside ChatGPT, a new guideline for developers, and a set of tools for law enforcement. Together, they cover the user, the builder, and the investigator.

What OpenAI is actually deploying

Start with detection. OpenAI has integrated external image classifiers into its systems, including the CSAM content classifier from Thorn, a nonprofit focused on child safety technology.

That classifier handles two separate jobs. The first is catching known CSAM through hash matching. The second is flagging new material, including AI-generated variants that have never been seen before.

The distinction matters more than it sounds. Hash matching works like a fingerprint database: an image gets reduced to a unique digital signature, and that signature is checked against a list of previously identified abuse material.

Advertisement

A freshly generated image has no fingerprint on file, so a system that only checks the list will wave it through. Novel-content classifiers exist to close that gap by judging what an image depicts rather than whether it has been catalogued.

The second piece lives inside ChatGPT itself. Users who try to create or access content that sexualizes children will now hit automated warnings, and repeat or serious attempts can lead to temporary blocks or account suspensions.

These interventions began a gradual worldwide rollout to personal accounts in mid-November 2026. They do not yet apply to enterprise accounts, educational accounts, or certain desktop interfaces.

Behind the warnings sits a broader layer of real-time classifiers that can process text, image, and audio content. The toolkit also includes context-aware classifiers and detection at the prompt level, meaning a request can be flagged based on what a user asks for, not only on what the model produces.

Developers get their own instructions. On November 15, 2026, OpenAI published CSAM guidance for API developers that walks through integrating perceptual hash matching alongside classifiers for novel content.

Perceptual hashing is the more forgiving cousin of standard hashing. Instead of requiring an exact match, it can recognize an image that has been resized, cropped, or lightly altered.

Finally, there is the reporting pipeline. OpenAI reports confirmed CSAM to the National Center for Missing & Exploited Children (NCMEC), the US clearinghouse that routes such reports to law enforcement.

Confirmed CSAM reports from OpenAI to NCMEC increased significantly in 2025, a rise the company attributes to the growth of its products.

How OpenAI got here

On April 8, 2026, OpenAI released a Child Safety Blueprint developed in collaboration with NCMEC and other organizations.

That document leaned heavily on policy. It emphasized that legal frameworks need to be modernized to deal with AI-generated CSAM, a category that existing statutes were not written with in mind.

Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.