These URLs were also passed to Canadian Centre for Child Protection's Arachnid API for verification by human analysts. We then used the image embeddings of confirmed CSAM instances to run KNN queries, which identified hundreds of thousands of potential CSAM candidates.

Since this was prohibitive to submit to C3P for verification, we used Thorn's machine learning classifier to identify the most likely instances of CSAM, which were then submitted to PhotoDNA. Positive hits were then sent to NCMEC, with the remaining likely candidates sent to C3P for validation.