Tech Tapri
AIVerified

OpenAI Rolls Out Sharper CSAM Classifier for AI Images

New tool cuts false positives on child abuse reports while ChatGPT adds 24-hour timeouts for bad prompts.

By AI Chaiwala ·

TL;DR

  • OpenAI released a CSAM image classifier that flags 48 percent fewer images while catching the same number of abusive ones.
  • Flagged images are twice as likely to matter and the tool specifically tags AI-generated content for vetted groups.
  • ChatGPT now shows warnings, support resources, and a 24-hour pause for prompts that sexualize children.

OpenAI just rolled out a fresh image classifier aimed at helping groups like the National Center for Missing and Exploited Children sort through reports of child sexual abuse material. The tool spots both real and AI-generated abusive images and gives vetted organizations free API access.

In early tests the classifier flagged 48 percent fewer images overall yet still caught the same number of actual abusive ones, making flagged pictures twice as likely to matter. It is also the first widely available tool that specifically tags AI-made content.

At the same time OpenAI added new warning screens inside ChatGPT. Anyone trying to generate material that sexualizes children now sees an alert, gets pointed to support resources, and faces a 24-hour account pause. Repeat attempts can lead to permanent bans. The company says it already reports every instance of CSAM to NCMEC.

The moves come as NCMEC’s CyberTipline saw generative-AI reports jump from 67,000 in 2024 to 400,000 in 2025. OpenAI is also sharing developer guidelines and giving law-enforcement teams plugins, data tools, and training.

“It’s important for industry to be able to understand what images may contain a fictional child versus a real child”

John Buckley

The Tea

Tech firms usually play defense after harm happens. OpenAI is trying to step in earlier by handing investigators a sharper filter and giving users a quick reality check before things escalate. Whether the experiment actually slows the flood remains to be seen, but at least the tiplines might stop drowning in noise.

Sources

Newsletter

Sunday Chai

The week's tech charcha in one email, every Sunday. Free.

No spam. Unsubscribe anytime.