Yeah, I do agree that in this instance you're probably correct. That said, it also isn't clear to me what in the tweet is easily mistaken for advocating self-harm - obviously keywords like "dying", "assisted suicide" and "failure" are probably playing a part but are not jointly sufficient. It could be that there's additional contextual information that could be seen as ideologically-valenced and that contributed to this moderation action, but we'll never really know.

For example, maybe the tweet caused a lot of harsh backlash for ideological reasons and that makes it more likely for Twitter to action a post for any reason, and the model is just making a softmax prediction of what that reason is. That's something that we should find discomforting.