Detect if the content contains toxic content from a paragraph of text

Comprehend is the best tool for this job.

You will want to create a Custom Classification where you can define words or phrases as "toxic". Unfortunately it's very hard to determine what constitutes "toxic" language out of the box as everyones definition is slightly different, so this will be the best path.

With Custom Classification, you can define words or phrases and define a label for them to train Comprehend and then run the specific comments through the Comprehend API to identify if the text contains any of your custom labels.