# Harassment Protection — Sylon

> Detects insults, threats and targeted harassment in any language — off by default, because it is your call to make.

Canonical page: https://sylon.app/features/harassment-protection

## Overview

A word filter catches the words you thought of. Harassment rarely uses them: it is spelled around, written in another language, or carried entirely by who it is aimed at. Harassment Protection reads the message and the conversation around it instead of matching a list, so it sees abuse a filter walks past. It is off by default and always will be — policing how members speak to each other is a decision a server makes deliberately, not one it inherits.

## How it works

1. **Reads the message, not a list** — Insults, threats and targeted abuse are recognised from what is written, in any language and any script. Deliberate misspellings and letter substitutions do not get past it the way they get past a word list.
2. **Context before conclusion** — The channel's own recent conversation is read alongside the message. Friends insulting each other in a room where that is the norm reads differently from the same words aimed at someone who was not part of it.
3. **Your threshold, your punishment** — Set how many detections a member accumulates before anything happens, and what happens then. Its own log channel keeps harassment cases out of the advertising log.

## Key capabilities

- Insult, threat and targeted-harassment detection
- Works in every language, without a word list to maintain
- Resistant to deliberate misspelling and character substitution
- Judged against the channel's own conversation, not a fixed standard
- Its own punishment, threshold and log channel
- Exempt roles, members and channels

## Frequently asked questions

### Why is this off by default?
Because deciding how members may speak to each other belongs to the server, not to the bot. Sylon was previously doing this as an undocumented side effect of Ad Protection; giving it a switch of its own is what turns it into an offer you can decline.

### Should I use this or the Word Filter?
They solve different problems. The Word Filter is exact and predictable — you list the terms, and those terms are blocked. This is judgement-based and catches what a list cannot anticipate. Many servers run both.

### What about friends who insult each other for fun?
Sylon reads the channel's recent conversation before deciding, so a room where that is the norm is judged as one. It is not perfect at this, which is why a timeout is a better starting punishment than a ban, and why every block can be undone in a click.

## Related features

- https://sylon.app/features/ai-ad-protection
- https://sylon.app/features/word-filter
- https://sylon.app/features/moderation

Keywords: discord harassment protection, ai insult detection, discord threat detection, anti toxicity discord bot, multilingual moderation