[Edit] And how should this be addressed in the US? Simple. Only censor things that are explicitly and clearly illegal under federal law and the laws of the state the company is headquartered in. If another state has an issue with that, then they must create their own competing platform.
Machine learning, what people today are mistakenly calling "AI" must tag anything that is not explicitly illegal with unobtrusive labels such as "hate speech" or "propaganda" or "misinformation" and so on.... Each label must link to the Machine Learning "Showing Its Work" to describe in detail exactly how it reached that conclusion and the members of the platform must have a way to contest and override the labels. The posts must show how many people disagree with said labels. No hidden algorithms, no hidden data-sets. The Machine Learning must show its work and show its references.
Members must have the ability to hide any content that has labels they do not wish to see just like the capabilities in StackExchange.