Stop Discord Crypto Scams with Anti-Scam Bot
Summary
CommunityOne Builders Help introduces a powerful, free anti-scam bot designed to secure Discord servers against evolving threats. The tool provides real-time monitoring for malicious links, fake investment images, and suspicious domains. By automating moderation, it helps server owners maintain a safe environment without manual effort, ensuring members are protected from phishing and crypto-related fraud.
Automatically detect and delete crypto scam content before your members can fall for it. Built for Discord servers that need zero-hassle protection against evolving scams.
The bot continuously monitors chat in real-time for:
Scam images — catches and removes fake investment ticket screenshots and proof images
Suspicious links — cross-references against known malware, phishing, and scam domain databases, following shortlink redirects before deciding
Disposable / young domains — flags and removes links and image-linked domains newer than a configurable age threshold
Invite link control — optionally block all Discord server invites (with permission bypass for mods)
Link whitelist — allow specific trusted domains like YouTube while removing everything else
Per-server and per-user language settings let the bot reply in whichever language fits your community.
Completely free, easy to set up, and actively maintained — new scam patterns are added as they appear.
**You can invite the bot here: **
https://discord.com/oauth2/authorize?client_id=1496505335597371504&permissions=1100048574531&integration_type=0&scope=bot
Support server: Join on Discord
Github: https://github.com/Mikael/antiscambot
Automatically detect and delete crypto scam content before your members can fall for it. Built for Discord servers that need zero-hassle protection against evolving scams.
The bot continuously monitors chat in real-time for:
Scam images — catches and removes fake investment ticket screenshots and proof images
Suspicious links — cross-references against known malware, phishing, and scam domain databases, following shortlink redirects before deciding
Disposable / young domains — flags and removes links and image-linked domains newer than a configurable age threshold
Invite link control — optionally block all Discord server invites (with permission bypass for mods)
Link whitelist — allow specific trusted domains like YouTube while removing everything else
Per-server and per-user language settings let the bot reply in whichever language fits your community.
Completely free, easy to set up, and actively maintained — new scam patterns are added as they appear.
**You can invite the bot here: **
https://discord.com/oauth2/authorize?client_id=1496505335597371504&permissions=1100048574531&integration_type=0&scope=bot
https://communityone.io/servers/1496506188597170236/antiscambot/
Github: https://github.com/Mikael/antiscambot
btw are you just getting rid of bad messages or you also handle user sanctions?
It does both, but everything is toggleable, by default its turned off
So bot can log, log and delete, mute, ban and dm the user
if you were referring to that
Changelog 5/22/2026
Security Improvements
- Added protection against oversized images that could overload the server — oversized images are safely ignored before being processed.
- Added phishing payload detection — suspicious file patterns in images are now blocked.
- Added image dimension limits to prevent memory exhaustion attacks.
Performance Improvements
- The image-scanning system now automatically adjusts how fast it processes images. When scam traffic spikes, it speeds up. When it's quiet, it slows down to save resources. Users shouldn't notice any difference, but server load is much more stable.
btw, have u thought about replacing OCR with gemma? Gemma is really good for moderation and it's easy to download and host urself
intereseting O: how do you detect phishing payloads?
gemma is also decent at image understanding tasks, but it does take some ram/vram...
I've tested 26b for this and it was very decent
Yeah, im trying it out
Every image attachment enters ImageScanService.scan_bytes, before opening the image or OCR runs, it does two security checks:
rejects empty/tiny payloads under 8 bytes as invalid_image_format, it calls _detect_suspicious_image_patterns, as shown below
def _detect_suspicious_image_patterns(self, payload: bytes) -> bool:
try:
check_bytes = payload[:64].decode("latin-1", errors="ignore").lower()
suspicious = ["<?php", "<%", "#!", "script>", "eval(", "exec(", "system("]
return any(p in check_bytes for p in suspicious)
except Exception:
return False
Sending a example in dm
wait, doesnt that mean that
fwiw I found this model recently https://huggingface.co/numind/NuExtract3, I haven't tried it yet but looks promising for doc parsing / moderation
specially since it's a small model
Have you found tesseract to be reliable? O:
Seems to be doing fine so far - I haven't heard any inaccurate hits yet
Awesome, i will check it out - thank you
AntiScamBot Patch Note
We’ve upgraded how AntiScamBot detects scam images.
The bot now uses a newer image understanding system instead of the previous OCR-based approach. This means AntiScamBot no longer has to rely mainly on pulling text out of an image and matching it against rules. It can now look at the full image context more directly, including things like fake Nitro gifts, suspicious Discord-style screenshots, fake promotions, scam links, and misleading visual layouts.
What changed
- Improved scam image detection.
- Better understanding of fake Discord/Nitro screenshots.
- Fewer missed scams caused by hard-to-read or stylised text.
- Fewer false positives from harmless screenshots.
- No more public “scanning” or benchmark-style messages in chat.
- If a scam image is detected, the bot silently follows the server’s configured actions:
- delete the message if enabled
- timeout the user if enabled
- send moderator alerts if enabled
- report according to server settings
Why this is better
The old system depended heavily on OCR, which meant detection could be affected by blurry text, weird fonts, screenshots, image compression, or partial text. Scam images often try to hide behind exactly those tricks.
The new system is better at judging the image as a whole, not just reading words from it. That makes it more reliable for modern scam posts, especially fake Discord gift screens, Nitro scams, impersonation attempts, phishing pages, and copied giveaway images.
What server owners need to know
Your existing moderation settings still apply. If your server has auto-delete, alerts, timeouts, or reports configured, AntiScamBot will continue using those settings..
AntiScamBOT Update
We’ve improved image scam detection to reduce false positives and catch more dangerous posts.
- Normal in-game screenshots, such as spins, tickets, rewards, inventories, and game UI, should be handled better.
- Posts telling users to turn off Windows Smart App Control, antivirus, Defender, firewall, or other security protections will now be treated as dangerous.
- The bot should be better at telling the difference between harmless game content and scam/malware instructions.
This update helps keep the server safer while reducing unnecessary flags on legitimate screenshots.
AntiScamBOT Update
New Join‑Account Scam Detection
- Checks newly joined members for scam/fake‑account signals using AI image analysis + profile metadata (username, avatar, bio, banner, badges, account age, etc.).
- Configurable via /setupbot and /antiscam‑settings: Off (default), Flag only, or Flag + Kick.
- When enabled and a suspicious join is detected, the bot posts an alert to the configured moderation channel.
- In Flag + Kick mode, the bot will attempt to kick the user.
Enable the feature in your server with /setupbot or tweak it later via /antiscam‑settings.
Patch Notes – AntiScamBot Update (2026-06-03)
What’s new:
- Longer watch for new members: The bot now monitors members for 24 hours after they join (up from 30 minutes), checking their messages throughout that period for scam patterns.
- Smarter scam detection for recent joins: Updated the model to look for specific scams like:
- Fake business offers (“inbox me for paid reputation/review removal”).
- Out-of-context funds/sales/profit screenshots used as proof of earnings.
- Broad creative/marketing bios (only flagged unless paired with stronger evidence like links or payment requests).
- Timeouts no longer trigger global watching: Being timed out in Discord (for any reason) will no longer cause the bot to watch that user globally for future messages.
- Cross‑guild checks are scam‑focused: When the bot sees a user who was moderated in another server, it only counts that as a risk if the prior action was scam‑related (not just a timeout or unrelated moderation).
- Faster image handling: The bot can now reprocess multiple images at the same time (using a thread pool), so it stays responsive while scanning pictures.
Patch Notes — June 22
Improved False Positive Handling
We’ve upgraded how AntiScamBot handles false-positive reports from server staff.
When staff click False Positive on a moderation alert, the bot now treats that report as a server-specific correction instead of changing detection globally.
What changed
- False-positive image reports now apply to your server only.
- If your staff mark an image detection as a false positive, AntiScamBot will avoid acting on that same image again in your server.
- This prevents one server from weakening protection for everyone else.
- If multiple independent servers report the same detection as a false positive, it will be sent to bot staff for manual review.
- Global detection behavior will not be changed automatically.
Abuse protection
To keep the bot safe:
- Only server staff can review moderation alerts.
- False-positive reports are counted by unique servers, not repeated clicks.
- One server cannot disable detections globally.
- Bot staff must review repeated cross-server false-positive reports before any global changes are made.
Why this matters
This makes AntiScamBot better at adapting to each community while keeping scam protection strong across all servers.
Your server gets more control over mistakes, but malicious servers cannot abuse the system to make scams harder to detect globally.
oh thatsss a pretty good idea, do users just use a command with the message link for the false positive?
That's an interesting approach, how do you think about moderation on images being server specific?
Like for example, I think about moderation in terms of categories. And a message may belong or not to a given category, so it's up to the server staff to decide which categories or types of behaviour they want to moderate.
So tweaking moderation to fit a server's style is a bit of an intriguing problem
uh, pretty much, if you click false positive on a lot of.. Roblox stuff, it will alert us, and if it’s a lot of staff from that server that clicked false positive, it will adjust its settings for that server, but not globally yet, for globally, x amount of servers needs to click false positive
then, it flags to bot staff saying x amount of servers clicked false positive on these stuff, check it out.
Oh I see
I was thinking on something like that, but instead of tweaking it at runtime automatically I would schedule time to manually review the false positives and adjust our detection.
We have a set of tests for each detection mechanism so I would add some generic of these to our testing dataset
I feel like the sweetspot must be somewhere in between our approaches, in particular our weakness here is that it does take time for us to update our detection
yeah that’ also works
AntiScamBot Update
- Better scam link detection: the bot can now follow redirects and short links to check where links really lead.
- Stronger protection from fake login/reward pages, including Discord/Nitro, Steam, Roblox, wallet, and giveaway scams.
- Added link reputation checks using trusted threat databases.
- Improved image scam detection: images can now be checked with OCR, meaning the bot can read text inside screenshots before using AI vision.
- More accurate decisions: the bot looks for multiple scam signals before taking action, helping reduce false positives.
- Better detection of suspicious domains, fake brand domains, and risky link patterns.
This update is focused on catching more real scams while staying careful with normal messages.
huh.. this one is hard.. 7 days is too aggresive... btw, automod allows u to ban based on account age already
definitivly check out sapphire
they are the new mee6
btw, have u thought about distribution strategies yet
Looking into a strategy
it seems like they’re using ocr and a domain list
idk, seems expensive to run a gpu server for that many servers for free
I once made some calculation for our own bot, with the way we analyze stuff with AI for moderation we can basically check 33.3k messages (30µUSD per message) per dollar, this estimation included messages with images iirc.
This with gemma 26b
oh fo sure
is the model you sent better, or this?
what do you mean with the "model I sent"?
Nuextract thingy
i see
