Security

What these bots are for

Security is not moderation. Moderation governs what members are allowed to say. Security governs who gets in at all, and what happens when somebody with elevated permissions does damage, whether deliberately or because their account was taken over.

Three jobs dominate. Verification establishes that a human is behind a join, via a button, a puzzle or a confirmation on a web page. Raid protection notices when an unusual number of accounts arrive in a short window and closes the server temporarily. Anti-nuke watches high-impact actions (mass bans, deleted channels, freshly granted roles) and strips permissions from whoever is performing them before they finish.

Features that actually matter

The thresholds have to be yours. Raid protection with fixed limits is either too twitchy for your server or too slow. Twenty joins a day and two thousand joins a day call for different numbers. Check that both the window and the count are configurable, and find out exactly what happens when the threshold is crossed.

Anti-nuke needs a memory. The whole value of the feature rests on telling a routine cleanup apart from an attack. Good implementations count actions per person per window and let you exempt known bots and roles. Poor ones fire the moment somebody renames three channels.

Verification without a dead end. Every hurdle at the door costs you members. If verification requires an external website, you lose everyone on mobile who cannot be bothered to switch to a browser. A button inside the server is almost always the better trade; it stops bots without driving people away.

Recovery. After an incident, what counts is how fast the previous state returns. Some bots back up roles and channel permissions and can restore them. It is the feature nobody thinks about until they need it.

Permissions worth checking

Security bots are the one category where broad permissions can genuinely be justified, and the principle still holds: only what the advertised feature actually needs.

Verification needs Manage Roles. Raid protection adds Kick Members and Moderate Members. Anti-nuke needs View Audit Log to see who did what in the first place, and Manage Roles to take permissions away from an attacker.

Anti-nuke bots in particular tend to ask for Administrator, arguing they must outrank everyone. The argument is not wrong, and it is also precisely why a compromised security bot is the worst case available. If you grant it, grant it only to a bot with a traceable origin, active maintenance, and a support server where questions get answered.

Role order matters as much as the permission list. The bot must sit above the roles it is expected to strip. Positioned below them, it will report the attack and be unable to act.

Where server owners go wrong

The first mistake is treating security as a one-time setup. Thresholds that fit a year ago do not fit a server that has since tripled. Review them twice a year.

The second is a single alert channel. If the alarm goes to the channel an attacker deletes first, you have no alarm. Keep a private channel that plays no part in day-to-day use.

The third is arming everything at once. Verification, raid protection and anti-nuke switched on together, without testing each in isolation, ends with nobody knowing which rule locked out whom. Bring them online one at a time.