Fighting online child sexual exploitation and abuse (CSEA) requires a collective defense across tech, law enforcement, and civil society. Protecting children is a core priority for Google, which is why we continuously invest significant resources into detecting, removing, and reporting this harmful content.
To keep pace with evolving abuse tactics and better support our partners, we are sharing an update on our latest technical innovations—including an industry-leading system to identify and tag AI-generated Child Sexual Abuse Material (CSAM), additional tools to support law enforcement agencies, and updated transparency metrics.
Industry-Leading GenAI CSAM Detection and Tagging
Google has a long history of evolving its tools to combat new threat vectors—such as when YouTube developed CSAI Match a decade ago as abuse shifted from images to video. Today, we are announcing the deployment of our industry-leading classification systems to identify and automatically tag AI-generated CSAM, which builds on our long legacy of detecting and reporting all forms of CSAM.
Reports to the National Center for Missing & Exploited Children (NCMEC) involving AI-generated content have surged, exceeding 182,000 in 2025 alone. Across the ecosystem, this sharp influx harms children, burdens reporting systems, and strains the resources of both analysts and law enforcement. Importantly, partners have shared that this additional manual step of determining whether content is authentic, synthetic or a combination can slow the response to children in urgent need of assistance.
To tackle this challenge, Google and YouTube are leveraging our AI classifiers to tag images at scale that contain CSAM generated wholly or partially with AI in reports to NCMEC, regardless of origin. This tagging is designed to provide additional context to a CyberTip for investigators without making a final determination of real-life vs. AI-generated harm.
In the last 30 days alone, this system:
- Processed reports and tagged roughly 50,000 files as likely AI-generated, representing ~6% of all files analyzed.
- Flagged at least one GenAI-tagged file per report, in approximately 11% of the 53,000 total reports sent to NCMEC.
As AI-generated child sexual abuse material continues to grow, tools like this help us respond more effectively to an increasingly complex threat. Being able to distinguish likely AI-generated content gives analysts and law enforcement valuable context, helping them prioritize cases involving real children who may be in immediate danger. We appreciate Google's continued investment in developing technology that supports faster victim identification and strengthens our collective effort to protect children.
Expanding Support for Law Enforcement
For over a decade, Google has been providing our Child Safety Toolkit free of charge to qualifying partner organizations to help them protect children by better prioritizing abusive content for review.
Now, new technological developments have allowed us to expand our collaboration with law enforcement agencies across the globe:
- Costa Rica Organismo de Investigación Judicial (OIJ): A new pilot implementation is expected to reduce forensic analysis times by 80% to 90% while limiting investigators' exposure to CSAM.
- Wyoming Internet Crimes Against Children (ICAC) Task Force: Using Google’s Content Safety API and open-source models, the task force is streamlining intake and triaging digital forensic evidence to locate victims faster.
- Thailand (TICAC): To help non-English-speaking agencies, in certain jurisdictions Google now automatically translates NCMEC CyberTips into local languages. In Thailand, this feature helped identify 47 victims from 43 translated reports this year.
According to Ryan Hieb, the Commander of the Wyoming ICAC Task Force, Google’s tools are “fundamentally transforming how the Wyoming ICAC triages CyberTips and digital forensics upon intake. It allows us to analyze digital forensics at an unprecedented scale, cutting through the noise to surface the most actionable intelligence. Ultimately, this partnership is accelerating our investigations so we can identify leads and recover child victims faster than ever before.’’
Transparency and Continuous Improvement
We recently updated our Global Transparency Report to differentiate between automated detection and manual reporting. Our proactive workflows—powered by hash-matching and machine learning—accounted for 99.57% of all content reported to NCMEC in the second half of 2025. This metric demonstrates our technological efficacy, as stopping this material before a user is ever exposed to it and before it spreads any further online remains our primary goal.