@kissane

(5/5) Getting back to the IFTAS issue, I'll point to the specifics of https://about.iftas.org/activities/moderation-as-a-service/content-classification-service/ again, because I often see "Come on!" replies that gloss over them:

• Their solution is summarized as "standardised processes and resources for monitoring distributed media", with hash databases curated by external entities

• "We cannot open the underlying hash matching databases, and any implementers will need to separately pursue access to these resources.": this is a clear constraint that server instances will hash media locally, and send hashes to "somewhere", for classification; this also implies that the classifying entity can build hash catalogs of all media present on the instances that request classification from it

• "If we find pertinent matches we’ll provide human review, notify you via email and issue a takedown request. We perform any required reporting and law enforcement record-keeping so you don’t have to." - Based on which jurisdiction?

• "CSAM" and "terrorism", two of the Four Horsemen of the Infocalypse, are used as a promotion tool for the hash-and-match approach; red flags all over here - "terrorism" as defined by Turkey, for the Kurds? By the PRC, for the Uyghurs? By Putin, for occupied territories? Again: who is curating the hash databases? Who is curating the curators? This is particularly troubling since automated reporting is already part of the system, and using NCMEC in a first step normalizes the reporting to authorities per se, by starting with a specific implementation that does not raise too many objections.

• "Once these hash databases are in place, we will explore text classification for issues like spam, child sexual exploitation, hate and harassment." - World-wide valid classification, based on what criteria?

It's important to realize that IFTAS being small and most likely not being a malicious actor is irrelevant. It does not take a decidedly malicious business organization like #Thorn (that colluded with EU commissioner #YlvaJohansson – there is ample material about hash-and-match as part of the EU's attempt at #ChatControl). All it takes is a useful door opener organization. I do not share your optimism that "currently ascendant systems are not inevitably annihilating forces, but legal and financial constructs that can be brought to heel, forcibly reconfigured, or just replaced" – especially in the age of grey intelligence, grey policing, and public-private partnerships.