Scam Detection
Automatically catch known scam images posted in your server
sushii can automatically catch messages that post known Discord scam images, like the "MrBeast" style deposit scams where images claim you can get a sum of money for free from an online casino promotion (often impersonating other influencers too). These scams spread without any links or text, so sushii will check only images for this feature.
What it does
When a member posts a message with one or multiple known scam images, sushii:
- Times the member out for 10 minutes.
- Deletes the messages.
- Posts an alert to your alerts channel with the image and quick Warn / Kick / Softban / Ban buttons.
- Logs it in your mod log, just like any other moderation action.
sushii only catches images it already knows about
This feature only works on scam images that have already shown up elsewhere and been confirmed as scams. A brand new scam image won't get caught the first time it appears anywhere.
False positives can happen
It's rare, but a normal image could occasionally look close enough to a known scam to get flagged by mistake. If that happens, please click the button in the alert message to report it as a false positive.
Setup
Set it up in /settings → Automod:
- Scam images: click Turn on to start detection. Off by default.
- Alerts channel is where alerts and the moderation buttons show up. Optional, but recommended.
- Exempt roles are roles that scam image and spam detection never touch (bots, trusted partners, etc).
sushii needs the Timeout Members and Manage Messages permissions to take action. If it can't time someone out, for example if their role outranks sushii's, it'll still delete the message and send the alert.
Where the scam images come from
sushii keeps a shared list of known scam images built from what's been seen across the publicly listed (Discoverable) servers it's in. This runs in the background all the time, whether or not you've turned the setting above on. This is what keeps the list useful for every server running sushii, including yours.
sushii does not scan every image it sees. It first looks for behavioral signals, such as the same images posted in several public channels within a couple of minutes, and only analyzes image content once that behavior crosses a high-confidence threshold.
sushii doesn't keep copies of your server's images. It stores a short fingerprint of each image, just enough to tell if two images match. You can't recreate or view the original image from that fingerprint. Images flagged as possible scams may be sent to an AI service to check whether they're scams.
See the Privacy Policy for the full details.