Dangerous content slipping through the cracks and the algorithms messing up is the same thing. There is no way for content to "slip through the cracks" other than via the algorithm.
You can view the content via direct links or search, recommendation algorithms isn't the only way to view it.
If you child porn that gets shared via direct links then that is bad even if nobody can see it, but it is much much worse if you start recommending that to people as well.
Everything is related. Search results are usually generated based on recommendations, and direct links usually influence recommendations, or include recommendations as related content.
It's rarely if ever going to be the case that there is some distinct unit of code called "the algorithm" that can be separated and considered legally distinct from the rest of the codebase.
