A product manager in China asks: can we find a way to prevent people commenting on these videos with comments that contain the word 共匪. Engineer goes away and comes up with a plan to train an AI model to detect these comments. He present's this solution to his manager who informs him that the solution needs to be in place next week, there is no time to gather the required training data and annotate it. There is also no budget for this project.
So the engineer goes off to find a new plan and finds that Google already has a basic spam comment detector which uses simple heuristics to delete comments. He adds a new rule to the detector. Mark as spam if it contains the "共匪" sub-string.
His boss is now happy, it gets sent to a test team in China who confirm that commenting with the word 共匪 on one of the videos results in the comment being deleted. Test passed job done, ticket closed. At no point does this get escalated or reviewed by anyone properly.
And if you don't believe this can happen anywhere, see the Scunthorpe problem https://en.wikipedia.org/wiki/Scunthorpe_problem