Possible Uses for LLMs

[TheTechromancer]

Right now there are a few tasks in BBOT that can't be automated in a traditional way, and I'm starting to think about how we could solve them with Large Language Models.

The task I'm thinking of right now is identifying different types of assets that are technically out of scope, but are still owned by the target and are extremely valuable for OSINT. The biggest example is mobile apps.

Say you want to find every mobile app published by your target, download and decompile them, and loot them for hardcoded secrets. We could automate plenty of the analysis with jadx/apktool, trufflehog, etc. But how would we get the APKs in the first place, without the risk of downloading random junk?

The human workflow for this is basically:

1. Based on intuition, infer the human-friendly name of your target (e.g. your target domain might be ea.com, but its human-friendly name would be "Electronic Arts")
2. Type the name into a search engine like apkpure.com.
3. Based on intuition, identify the apps in the search results that belong to your target.
4. Based on intuition, determine whether the publisher of the apps is the same as your target, and if so, also download all the apps belonging to that publisher.

Obviously this task relies a lot on intuition. But this is exactly the type of workflow that can be automated with LLMs using tools like langchain.

@domwhewell-sage