OpenAI unveils tool to better detect child sexual abuse material
Guardii Analysis
On November 16, 2026, OpenAI released a new classifier tool designed to detect child sexual abuse material, including AI-generated content. The company reported that the classifier flags approximately 48 percent fewer false positives than current detection tools while identifying a similar volume of abusive images, and that flagged images are about twice as likely to contain actual abuse. All flagged material requires human review for accuracy. OpenAI announced that it is also introducing a new in-product warning message in ChatGPT for users who attempt to request child sexual abuse material, which directs them to behavioral health resources and states plainly that sexualizing children causes harm even when the content is AI-generated.
OpenAI's detection classifier addresses the platform-side moderation challenge after material has been generated or uploaded. The Guardii platform, a Meta Business Partner backed by Startmate, operates upstream of this problem by detecting the coercion, grooming or sextortion conversation in real time as it unfolds in a child's direct messages on Instagram, Snapchat, Discord, Roblox and other services. Where a classifier scans completed images for signs of child sexual abuse material, Guardii's anti-grooming and anti-CSAM modules analyze the shape and escalation of a conversation to surface a child at risk to a parent or safeguarding authority before an offender successfully solicits imagery, real or AI-generated. This approach reduces the volume of material that platforms and law enforcement must later triage by blocking hostile or coercive contact before it reaches the child.
