
OpenAI Ships Free CSAM Classifier That Cuts False Flags by 48%
OpenAI's new free API classifier flags 48% fewer images while finding the same number of abusive ones, alongside ChatGPT warnings and new developer CSAM guidelines.
- By
- Sophie Lindqvist
- Filed
- Channel
- AI & Compute
- Read
- 3 min read
OpenAI released an image classifier today that flags 48% fewer images while detecting the same number of abusive pictures, and it is giving the tool away for free through its API to select companies.
The classifier detects child sexual abuse material (CSAM), including images generated by AI. It examines visual signs of abuse in previously unseen images and assigns a score. Reviewers use that score as an indicative signal when deciding how to handle flagged material. OpenAI also reported a second figure from early partner testing: images flagged by the classifier are twice as likely to be genuinely abusive, which cuts the review workload on both ends.
The company developed the tool with the National Center for Missing and Exploited Children (NCMEC). It is working with Tech Coalition and ROOST to share access. No pricing applies; access is free but limited to selected companies.
What changes inside ChatGPT?
ChatGPT now shows warning screens to users who attempt prompts involving the sexualization of children. The alert directs users toward specialized support intended to stop harmful behavior. The app also suspends the account for 24 hours as a first sanction. The system already blocks CSAM content and related requests. Repeated offenses can lead to permanent account removal, OpenAI said.
OpenAI did not publish detection rates for the warning system itself, so its deterrent effect remains unmeasured outside the company. The message design, however, has external grounding: OpenAI formed the warnings with organizations and researchers including Protect Children, the Lucy Faithfull Foundation, and Joel Scanlan at the University of Tasmania, who independently evaluated a trial of deterrent messages.
What do developers and police get?
The company published guidelines for developers building on OpenAI tools, covering how to prevent, detect and help users report CSAM. In early testing, OpenAI said the guidelines helped a number of developers adopt child safety tooling and register to report CSAM to NCMEC. The company gave no count of developers, so adoption scale is unverifiable.
Law enforcement agencies handling child safety cases receive a separate package:
- Dedicated plugins for investigations
- AI-powered investigative tools
- Complex data analysis features
- Access managed by program administrators
- Training for investigators plus a field guide on using AI
Why now?
The release follows sustained legal and reputational pressure. Several lawsuits blame OpenAI's technology for harming users, especially children. Safety-focused nonprofit Common Sense Media labeled OpenAI's tech an "unacceptable risk" for teens. Competing chatbots have not fared better; Grok has been used to create nonconsensual sexualized imagery, including of children.
Over the past 12 months OpenAI has released a safety blueprint, updated content guidelines to bar inappropriate content, retuned its model to resist generating harmful content, and shipped a special version of ChatGPT for teens. Today's tools extend that track from policy into measurable detection performance.
The 48% figure and the doubling of flag precision come from OpenAI's own early partner testing, not from an independent audit. If those numbers hold at scale across the API partners receiving free access, the classifier could become a de facto industry screening standard; if they do not, OpenAI will face fresh questions about how it validates the safety claims central to its legal defense.
Source: techcrunch.com
More from Sophie Lindqvist
Show full bio
News editor covering money, funding and market shifts around Semiconductors, chip manufacturing, hardware, AI compute at Chip Dispatch.
1,052 articles
Related articles
openai-releases-free-csam-classifier-built-with-ncmec
OpenAI Releases Free CSAM Classifier Built With NCMEC
openai-s-csam-classifier-flags-48-fewer-images-catches-twice-the-abuse
OpenAI's CSAM Classifier Flags 48% Fewer Images, Catches Twice the Abuse

