This is a really weird thing to do, and a really weird basis for complaint about ML recommenders. I'd imagine that 99.9% of people who watched _30 minutes_ of an artsy foreign-language noir film are people who are inclined to enjoy artsy noir films (or at least trying it out). The only reasonable complaint here is that recommender systems can be overly sticky when they don't provide a way to provide strong negative feedback (eg YouTube's "not interested" option).
If you have 100 movies on your history, and one is an artsy one that you got away after 30 minutes, the evidence is pointing that you will not like anything else like it.
But this does appear to be a minority position.
For example, in a scene the main character might see a new person for the first time. There's nothing in the movie so far to indicate that the main character knows the other person. Yet the subtitles may reveal that other characters name when the subtitled dialog goes "CARL: Hello there", which might ruin an important plot point.
I really don't understand why they keep doing this.