- The new classifier cuts false-flagged images by 48% while roughly doubling precision on flagged content
- Vetted organizations get free API access; NCMEC supported training, evaluation and deployment
- ChatGPT interrupts related requests with a 24-hour suspension, escalating to permanent bans or law-enforcement referral
AI CURATED HOT STORY · 11/17/2026
OpenAI Launches CSAM Detection Classifier with 48% Fewer False Positives
OpenAI has released an image classifier that identifies child sexual abuse material (CSAM) and determines whether it is AI-generated, making it freely available via API to vetted organizations. The tool significantly improves accuracy, reducing false positives by 48% compared to existing measures and doubling the likelihood that flagged images actually contain abusive content.
The classifier was trained, evaluated, and deployed with participation from the National Center for Missing & Exploited Children (NCMEC). OpenAI also provides detection guidance for developers and training for law enforcement. ChatGPT now includes an intervention mechanism: requests involving CSAM are interrupted, with a 24-hour temporary ban that escalates to permanent suspension or referral to law enforcement for repeat offenders. Additionally, OpenAI, Anthropic, and Google jointly commissioned Tech Coalition to publish industry handling guidelines.
The move comes as generative AI-related reports to NCMEC's CyberTipline surged from 67,000 in 2024 to 400,000 in 2025, putting mounting pressure on platforms. OpenAI acknowledges the collaboration is still experimental and aimed at prevention, but the free availability of the tool and the release of industry guidelines mark a significant step toward collaborative, standardized content safety governance, potentially driving unified CSAM detection and response standards across the industry.
Hourly heat curve
10 fully observed hoursThe line compares only sources observed throughout; gap hours are interpolated to keep the trend continuous.
Coverage timeline
Total 2 reports · Latest first- IT之家 · 智能时代T2·58 pts
- The Verge AIT2·62 pts
- The classifier flags child abuse imagery and whether it is AI-generated, with free access for vetted organizations
- NCMEC's CyberTipline saw generative AI-related reports jump from 67,000 to 400,000 in a year
- OpenAI adds developer guidance and law enforcement training, though it calls the effort still experimental