If this even comes close to doing any “recognition” on local content and isn’t just (as the article says) a simple lookup of known images through hashes, then that would be a massive scandal. So I’m giving them the benefit of the doubt until it’s clarified, as the only confirmed way it works is the image hash lookup.

There are words like “neural” and “trained” here, but no clear sign that there is actually any kind of image recognition that would risk false positives other than hash collisions.