If Facebook repeatedly permits copyright infringement on the same video, they might get sued by the rightsholder, which might be a very wealthy company with experience enforcing its rights in courts.

In they repeatedly make a $30,000/year employee view 15 seconds of the same child being beaten to a pulp (...the minimum time they are obligated to watch each clip before rendering a judgment), the employee will just quit after their health declines past a certain point.

So while the technology could be applied to both contexts, there's probably a lot more economic incentive to apply it to the first context than the second one.

Machine learning for recognizing duplicates is only for "protection of IP rights", not for protection of low wage workers. Apparently somebody at facebook decided that ML is more expensive than low wage human worker rejecting the same material over and over. Layers are more expensive and bring more bad media coverage than someones suicide in 3rd world country.