OpenAI Adds Child Abuse Content Detection to ChatGPT
The company is rolling out a classifier from Thorn and automated warnings to block attempts to create or access child sexual abuse material.
Updated on Nov. 16, 2026 in Artificial Intelligence

OpenAI has integrated an image classifier from Thorn that detects known and new child sexual abuse material, and ChatGPT now issues automated warnings to users who try to create or access content sexualizing children. Repeat or serious attempts can result in temporary blocks or account suspensions.
Why it matters
Hash matching can only catch previously identified abuse material, so classifiers that flag new content, including AI-generated variants, address a gap in detection. The changes also set expectations for how AI platforms report confirmed material to the National Center for Missing & Exploited Children.
The classifier reduces an image to a unique digital signature checked against a list of previously identified abuse material, and uses perceptual hashing to recognize resized, cropped, or lightly altered images. OpenAI has not published figures on how many reports it has filed with NCMEC in 2025 or how many users have triggered warnings.
The players
OpenAI
OpenAI is an artificial intelligence research and deployment company that develops the ChatGPT chatbot and the API used by developers.
Thorn
Thorn is a nonprofit organization that builds technology to defend children from sexual abuse, including tools for detecting abuse material online.
National Center for Missing & Exploited Children (NCMEC)
NCMEC is a U.S. nonprofit that operates the national clearinghouse for reports of child sexual exploitation and works with technology companies on detection.
The details
The classifier, supplied by Thorn, performs hash matching against known material and flags new content, including AI-generated variants. Users who attempt to create or access sexualized content involving children receive automated warnings in ChatGPT, and the rollout to personal accounts began gradually worldwide in mid-November 2026.
Timeline
April 8, 2026: OpenAI released its Child Safety Blueprint with NCMEC.
November 15, 2026: OpenAI published CSAM guidance for API developers.
Mid-November 2026: ChatGPT warnings began a gradual worldwide rollout to personal accounts.
The Tech Race
The move extends the hash-sharing model that NCMEC's CyberTipline has relied on for years to AI-generated content, which earlier detection systems could not identify. It positions OpenAI alongside other platforms working to close the gap that novel imagery creates for reporting pipelines.
Users who attempt to generate or search for sexualized content involving children will now see automated warnings in ChatGPT, and repeated attempts can lead to temporary blocks or suspensions. Enterprise, educational, and some desktop accounts are not yet covered by these interventions.
The takeaway
Account-level warnings and blocks now sit alongside hash-based detection, so users who run into these limits will be notified in the product itself. Organizations building on OpenAI's API should review the November 15, 2026 developer guidance on CSAM.
Further reading
For more on how AI platforms are handling safety tooling, see Artificial Intelligence coverage.



