There are several harm factors which are hidden from plain view.

Commonly used hash algorithms are too rigid. A detection system shouldn't be bypassed just when a single bit gets flipped in a large file. The algorithm need to be fault tolerant, give few hash collision, and be proven by the passage of time. This of course a contradiction in terms.

It also need to be maintained and safe guarded against abuse. Who will watch the watchers, and how do we control what gets defined as offending images if there is no public reviews?

How is legal rights handled? How should appeals be handled, and peoples right to face ones accuser.

How do we control scope creep so "unwanted" political competition don't get suppressed under anti-propaganda laws? How is the slippery slope argument handled?

And last... but far from least, is the classical argument of 20th century political environment: Practicality. What does the cost-benefit analyses say about such filters and databases. Is the maintenance that those databases require cost more than they provide to society? What other options has been thought of, and how does the databases compare in efficiency and cost?

So if we think about it, maybe the harms are not that well hidden.

The problem is, that they have then equipment to monitor web traffic at the content level. And it is nowhere clear, if the hashes are actually offending images, either maliciously or because the database is just on a similar quality level as the various databases for copyright violations. [1]

[1]https://www.techdirt.com/articles/20120806/11053019945/curio...

Actually child porn images are probably much easier to detect than YouTube blocking videos. One aspect of copyright infringement that makes content detection hard is that there is fair use. There is absolutely no fair use for child porn.

And that database never has false positives?