Online social-media platforms are a vital arena in which socio-political perspectives are put forth, debated, and spread. Prior work suggests that certain moral-emotional language drives online contagion, but theoretical and empirical findings remain debated, inhibiting constructive interventions. We substantially advance this ongoing debate using a diverse range of topics and both mainstream and extremist social media platforms. We find two countervailing dynamics predict the rise and fall of posting engagement online. First, we confirm that content infused with a greater number of moral words is associated with increased engagement; still, contrasting with prior work, we find no evidence that it is driven specifically by moral-emotional words as opposed to the more general and larger lexicon of moral language. Second, we identify a striking reversal in the relationship between moral language and engagement, a phenomenon we call “moral penalty.” We find that as the ratio of moral to non-moral words surpasses a threshold, the process of engagement around a post reverses with marked decreases in engagement and online diffusion. These findings help clarify links between morality and online engagement and contagion: infusing messages with moral language increases their spread to a point after which embedding posts in morality may backfire.