What Is AI Moderation, and Does Discord Need It?

Keyword lists, pattern matching, and rate limits catch a lot — but they're built to catch what someone anticipated in advance. A message that's ambiguous, uses phrasing nobody thought to block, or combines several small red flags that individually wouldn't trip anything, can slip through a purely rule-based system.

Discord's native AutoMod evaluates message content directly against rules you define — keyword lists, a preset spam filter, mention-spam limits. It has no concept of reviewing a genuinely ambiguous case with broader context; a message either matches a rule or it doesn't.

Deterministic signals run first — AI Moderation only reviews candidates those signals already flagged as worth a second look: scam-like links, configured phrases, spam, mass mentions, suspicious attachments, or activity in designated high-risk channels.

• Covers 14 categories: violence and threats; scams, fraud, and illegal activity; sexual harm; child safety; extortion; high-impact advice; privacy exposure; piracy; dangerous weapons; hate or harassment; self-harm; sexual content; election misinformation; and malware or cyber abuse.
• Choose Alert staff only for an advisory workflow, or Delete confirmed unsafe messages for confirmed destructive action — with optional escalation requiring stronger confirmation on serious or ambiguous results before deletion.
• Configure a private alert channel, the staff roles allowed to use alert actions, candidate phrases, high-risk channels, ignored channels or roles, and scoring thresholds.
• It never bans, kicks, or times out members — a delete action still requires the configured confirmation path and Discord permission.

1. Confirm Server Premium is active — AI Moderation is opt-in and Premium-only.
2. Open Moderation in the dashboard and enable AI Moderation.
3. Choose the safety categories relevant to your community — you don't have to enable all 14 at once.
4. Set candidate phrases, high-risk channels, and any ignored channels or roles.
5. Choose Alert staff only to start, review real results for a while, and only move to Delete confirmed unsafe messages once you trust the configuration.

• Expecting it to ban, kick, or time out — it never does; pair it with /mod or Warning Points for destructive action.
• Enabling Delete mode on day one — start with Alert staff only so you can validate accuracy before anything gets removed automatically.
• Treating a flag as certain — automated safety analysis can be wrong or lack context; keep staff review and appeal procedures appropriate to your community.
• Enabling all 14 categories without considering which are actually relevant to your community's real risk profile.