discord-scam-detection

Experimental detection, and proof-of-concept, of Discord Nitro phishing/scam links via analyzing images.

This repository is not considered production-ready. Use with caution.

Setup

Set environment variable BOT_TOKEN to your bot's OAuth 2 token and run.

Background

Initial thoughts

A few days ago (January 2022), I had a thought of detecting Nitro phishing links by analyzing the image in the message embed. Most Nitro scam websites use the same, or similar, images, and all of them are Discord's images (some images even link to Discord's website directly, and aren't even self-hosted.) I spent about 2 hours total making this small project to test out my theory, and after testing against two scam websites has proven effective.

Method of detection

This works by analyzing the image of a URL, if the URL has an embedded image. Malicious image hashes are stored in image_analyzer.go. The hashes are hex-encoded SHA-256 hashes of the image content. If the URL is discord.com, the message is obviously ignored for detection since the image(s) will be legitimate.

Why this method instead of pre-existing methods?

This method of detection is much more effective than checking against a list of domains, such as this repo, since scammers can easily register a new domain for only a few dollars (or even less) and host a new website for free. In this case, any manually-created list of domains becomes ineffective until manually updated again; which could take several hours, or even days, before the new domain becomes widespread.

Obviously, this method can also be circumvented by using different images that are not listed in the list of hashes. This method can also be circumvented by changing a single pixel in the image, since the entire SHA-256 hash would change. However, this is much more effective than the aforementioned method of checking against a list of domains, and is future-proof against new domains (assuming new websites use images in the list of malicious hashes.)