Skip to content
VECTOR WIREAI INTELLIGENCE
UTC
Home

OpenAI ships CSAM classifier that halves reviewer workload

48% fewer flagged images with equal abuse detection in early partner testing.

OpenAI announced an image classifier designed to detect child sexual abuse material, including AI-generated content, and is offering vetted organizations free API access to the tool1,2.

The classifier was developed with the National Center for Missing and Exploited Children (NCMEC), and OpenAI is working with Tech Coalition and ROOST to distribute access. In early testing with partners, the classifier flagged 48% fewer images while identifying the same number of abusive images, cutting reviewer workload; images the tool did flag were twice as likely to be actual abusive content.

Beyond the classifier, OpenAI is adding warning screens inside ChatGPT to block users from accessing material that sexualizes children. Accounts that attempt to access such content face 24-hour suspensions4. The company is also publishing new guidance for developers on detecting this category of content and helping train law enforcement on how AI can aid investigations.

OpenAI disclosed the tools at a Monday press briefing in Washington, DC. The announcement arrives as policymakers push for stronger safety restrictions on AI tools.

Separately, Britain's media regulator Ofcom announced new online child-safety rules on November 16, requiring platforms to ban users who generate, upload, or share CSAM and to expand use of detection technology, including automated tools for previously unknown content3. Those rules are expected to take effect from early 2027, subject to parliamentary approval.

ANALYSIS OpenAI's decision to make the classifier available at no cost through its API positions the tool as shared infrastructure rather than a proprietary safeguard, a model that could lower adoption barriers for smaller trust-and-safety teams.

The moves follow OpenAI's recent step to embed four independent safety assessors inside its development pipeline[2], broadening the company's public safety commitments beyond model capability risks to content-level harms.

Was this useful?

The day’s vector, in your inbox.

Vector Wire Daily: one dispatch each publishing day.