Enter Sandbox 27: Account creation

It’s been nearly 4 years since I published my last article in this series providing the community with a large corpora of sandbox reports (apilog_2019-07-14).

One of the less known (but still pretty interesting, artifact-wise) findings inside this 200MB+ file is a large number of “net user” command invocations that can be attributed to either reconnaissance activities or are basic account creation commands issued by malicious samples…

Yes, when you look at these invocations on an individual sample level they probably don’t stand out too much and don’t have much of an impact, but if you look at them in bulk, they do light up the terminal like a xmas tree…

Let’s have a look!

We execute a ripgrep search on the apilog_2019-07-14 file:

rg -i --iglob apilog_2019-07-14 "net\s+user"

– it is looking for any ‘net user’ case-insensitive string references preserved within this large file. The results give us a lot to think about… but, more importantly, force us to do some time-consuming manual clean up first ;), then finally leading us to the following list

Do you like what you see?

Not installing the installers, part 2

In the last post I described how we can pull some interesting metadata from decompiled installers. Today I want to discuss one practical example of how this data can enrich analysis, both manual and automatic (f.ex. sandbox, EDR).

Many programs cannot be properly analyzed by sandboxes, because they require command line arguments. While command line options for native Windows OS binaries are usually well documented (well, not really, there is a lot of undocumented stuff, but let’s forget about it for a second), command line options used by goodware is a completely different story. And of course, even worse for malware.

The good news about goodware is that they handle command line arguments in a very predictable way. The string comparisons are usually ‘naive’, direct and not optimized, and often, the programs include actual help that can be seen after running the program with the /?, /help, -h, –help arguments. And very often, a search for ‘usage’ keyword inside the binary can help us to find the options that program recognizes. f.ex. this is what we see inside cscript.exe:

Predictable is good, and can serve at least a few purposes:

  • we can generate a list of known parameter strings that goodware typically uses (and even attribute it to specific software vendors)
  • we can create yara signatures for these
  • we can incorporate this set into EDR command line parsing routines to assess invocation’s similarity to known good
  • we can also leverage this to run the sample in a sandbox or manually with these easily discovered command line arguments (kinda like assisted fuzzing)
  • we can include these findings in the report itself to hint analysts they may need to do some manual reversing (it would seem the program accepts the following command line arguments…)

Looking for typical command line arguments is actually quite difficult. There are a lot of ways to implement string comparisons and as I explained long time ago in one of my sandbox series, there are like gazillion different string functions out there. Plus different compilers, different optimizations make the code even harder to comprehend. Naive search for /[a-zA-Z_0-9]_/ could work on a binary level, but this is going to hit a lot of FPs. Decompiled scripts can come to the rescue, as they include actual invocations of programs and specify precisely what parameters will be passed to the program.

The attached list focuses on a basic command line argument extraction (just the /foo part) from around 10K decompiled scripts. More advanced analysis would include options taking parameters (f.ex. /foo bar).

You can download it from here.