Its inevitable in public spaces, especially when there is an ounce of anonymity.

you could use llama guard or openai's omni-moderation model to flag bad actors