June 5, 2023

Fighting spam: our new approach

In Q1 2023, automated bots uploaded over 2 million spam images to PixServe. These ranged from SEO spam and phishing lures to outright scam advertisements. The volume was unlike anything we'd dealt with before, and our existing manual review process couldn't keep up.

Over the past three months, we've built and deployed a multi-layered defense system. The first layer is aggressive rate limiting: we now cap uploads per IP address and throttle accounts that exhibit bot-like upload patterns. This alone blocked about 60% of the spam we were seeing.

The second layer uses perceptual hashing. When we identify a spam image, we generate a perceptual hash and add it to a blocklist. Any future upload that matches within a certain Hamming distance is automatically rejected. We've also integrated with NCMEC's hash-sharing program for automated CSAM detection — any match is immediately flagged, removed, and reported.

These measures are helping, but we're evaluating whether additional steps — including requiring account verification — may be necessary in the future. We'd rather keep PixServe as open as possible, but not at the cost of the platform being used for harm. We'll keep you updated as our approach evolves.