Introducing SiriusGPT: The First Real-Time GPT / LLM AI based Antimalware Solution

  • Thread starter Thread starter danb
  • Start date Start date
  • Featured

danb

From VoodooShield
Thread author
Verified
Top Poster
Developer
Well-known
Forum Veteran
May 31, 2017
2,069
12,014
2,978
Overland Park, KS
Hey Guys!

Developing SiriusGPT was by far the most difficult project I have ever worked on, but I am super happy with the results. SiriusGPT utilizes 3 LLM / GPT models in an iterative configuration, and I have yet to see one false positive or negative, for both compiled and text based files, and for command line analysis as well.

If you scan a file or experience a block, you can click the Sirius logo to toggle between the Analysis Report and the event details. I am pretty happy with the user prompt, but we will probably refine it a little more as we go, and if anyone has any suggestions, please let me know.

A lite version of CyberLock is implemented into SiriusGPT, so you can select from AutoPilot and Dynamic Security Postures (Smart Mode), which is essentially a zero-trust security posture when you are browsing the web or checking emails. The LLM command line / LOLBin / fileless analysis is fully implemented, so we no longer have to worry about confusing command line blocks, they are all handled automatically, unless there is a malicious attack, in which case you will see a prompt that tells you to not allow the event. There should be close to zero bugs, but as soon as we are sure all of the bugs are worked out, we will be implementing the SiriusGPT engine into our other products.

The LLM analysis on these events are extremely accurate, usually with a confidence of 95% or higher. Here is a sample analysis report for these types of events.

Process path: c:\windows\system32\rundll32.exe
Parent process path: c:\windows\explorer.exe
Command line: "c:\windows\system32\rundll32.exe" using s=system;class p{static void main(){s.console.writeline((char)72+""+(char)101+(char)108+(char)108+(char)111);}}

Final Verdict: Not Safe with 95% confidence.

## Analysis Summary

The given command line execution event involves `rundll32.exe` being launched by `explorer.exe`. The command line argument passed to `rundll32.exe` is unusual and appears to be an attempt to execute arbitrary code. `rundll32.exe` is a known LOLBin (Living Off The Land Binary) that can be used to execute code in a DLL by calling a specific function within it.

The command line argument `using s=system;class p{static void main(){s.console.writeline((char)72+""+(char)101+(char)108+(char)108+(char)111);}}` is not a typical DLL function call expected by `rundll32.exe`. Instead, it resembles C# code that prints "Hello" to the console. This syntax is not directly executable by `rundll32.exe`, which typically requires a DLL name and an entry point.

The use of `rundll32.exe` with such an argument is suspicious because it attempts to abuse the binary for executing arbitrary code, potentially bypassing security restrictions. The code within the command line argument is obfuscated or encoded, as it uses character codes to represent the string "Hello". This behavior is indicative of an attempt to hide the true intent of the command.

The parent process, `explorer.exe`, launching `rundll32.exe` with such an argument is abnormal. While `explorer.exe` can launch various processes, the context and the command line argument used here suggest malicious intent.

## Malware Type and Name

Given the suspicious use of `rundll32.exe` and the encoded command line argument, this activity represents LOLBin Abuse. The specific behavior of executing C#-like code within the command line argument is not typical for legitimate applications.

## Final Verdict

Malware type: LOLBin Abuse (Thanks to @Trident for the malware name idea!)
Malware name: LOLBin.Rundll32.Exec (Thanks to @Trident for the malware name idea!)
Final verdict: Malicious with 95% confidence.

If you are curious and want to try SiriusGPT, but have CyberLock installed, you can uninstall CyberLock temporarily and then install SiriusGPT. You might want to click No when it asks if you want to delete the Settings and Logs during the CyberLock uninstall.

You can download the installer or portable version here:




Thank you guys!


Dan
 
Named after the dog star constellation? The doggo icon suggest this anyway.
Hehehe, that did not last long... 7 minutes ;). Yes, my dog Gracie is very patient with me working 80+ hours a week on these projects, but we still always go to the park 2-3 times a day. And when we have some down time, I really make it up to her. But yeah, that is correct sir. Please email me the email address you want to assign to your new lifetime license. It can either be an existing or new account, whatever works for you.
 
@danb - Excellent work as always, I will download and install and have a jolly good play with it.
My dog Bob is the same he waits patiently for me to have breaks and take him for walks when I work at home... Bob is my best fried "don't tell my partner" he always bring a smile.
 
BTW, the first person who guesses why I chose the name Sirius gets a free lifetime license to CyberLock / SiriusGPT ;).
brightest star in the sky... Edit I must have overslept, @n8chavez beat me to it, congratulations. Uninstalling Cyberlock was semi-traumatic for me after all the years running VS -> CL but I guess SiriusGPT is the new CL, or however danb works out the implementation. So far so good. Also running with TrendMicro on this VM. No issues seen so far... :cool:
 
Last edited:
Malware type: LOLBin Abuse (Thanks to @Trident for the malware name idea!)
Malware name: LOLBin.Rundll32.Exec (Thanks to @Trident for the malware name idea!)
Final verdict: Malicious with 95% confidence.
I like contributing with ideas.

Anyway, depending on the test results I will either send private messages, or text here.

Let’s see if it’s mature enough to be reviewed.
 
Hey Guys!

Developing SiriusGPT was by far the most difficult project I have ever worked on, but I am super happy with the results. SiriusGPT utilizes 3 LLM / GPT models in an iterative configuration, and I have yet to see one false positive or negative, for both compiled and text based files, and for command line analysis as well.

If you scan a file or experience a block, you can click the Sirius logo to toggle between the Analysis Report and the event details. I am pretty happy with the user prompt, but we will probably refine it a little more as we go, and if anyone has any suggestions, please let me know.

A lite version of CyberLock is implemented into SiriusGPT, so you can select from AutoPilot and Dynamic Security Postures (Smart Mode), which is essentially a zero-trust security posture when you are browsing the web or checking emails. The LLM command line / LOLBin / fileless analysis is fully implemented, so we no longer have to worry about confusing command line blocks, they are all handled automatically, unless there is a malicious attack, in which case you will see a prompt that tells you to not allow the event. There should be close to zero bugs, but as soon as we are sure all of the bugs are worked out, we will be implementing the SiriusGPT engine into our other products.

The LLM analysis on these events are extremely accurate, usually with a confidence of 95% or higher. Here is a sample analysis report for these types of events.

Process path: c:\windows\system32\rundll32.exe
Parent process path: c:\windows\explorer.exe
Command line: "c:\windows\system32\rundll32.exe" using s=system;class p{static void main(){s.console.writeline((char)72+""+(char)101+(char)108+(char)108+(char)111);}}

Final Verdict: Not Safe with 95% confidence.

## Analysis Summary

The given command line execution event involves `rundll32.exe` being launched by `explorer.exe`. The command line argument passed to `rundll32.exe` is unusual and appears to be an attempt to execute arbitrary code. `rundll32.exe` is a known LOLBin (Living Off The Land Binary) that can be used to execute code in a DLL by calling a specific function within it.

The command line argument `using s=system;class p{static void main(){s.console.writeline((char)72+""+(char)101+(char)108+(char)108+(char)111);}}` is not a typical DLL function call expected by `rundll32.exe`. Instead, it resembles C# code that prints "Hello" to the console. This syntax is not directly executable by `rundll32.exe`, which typically requires a DLL name and an entry point.

The use of `rundll32.exe` with such an argument is suspicious because it attempts to abuse the binary for executing arbitrary code, potentially bypassing security restrictions. The code within the command line argument is obfuscated or encoded, as it uses character codes to represent the string "Hello". This behavior is indicative of an attempt to hide the true intent of the command.

The parent process, `explorer.exe`, launching `rundll32.exe` with such an argument is abnormal. While `explorer.exe` can launch various processes, the context and the command line argument used here suggest malicious intent.

## Malware Type and Name

Given the suspicious use of `rundll32.exe` and the encoded command line argument, this activity represents LOLBin Abuse. The specific behavior of executing C#-like code within the command line argument is not typical for legitimate applications.

## Final Verdict

Malware type: LOLBin Abuse (Thanks to @Trident for the malware name idea!)
Malware name: LOLBin.Rundll32.Exec (Thanks to @Trident for the malware name idea!)
Final verdict: Malicious with 95% confidence.

If you are curious and want to try SiriusGPT, but have CyberLock installed, you can uninstall CyberLock temporarily and then install SiriusGPT. You might want to click No when it asks if you want to delete the Settings and Logs during the CyberLock uninstall.

You can download the installer or portable version here:




Thank you guys!


Dan
Great, was looking forward to the release. Does it update automatically to newer versions? I always kind of miss a "check for updates" button in your products I have to admit.
 
Hey guys, I will catch up on the posts I missed later today, but I wanted to mention that we had some false positives for some command line blocks. I just now rewrote the instructions to the LLM's so that it understand that there are sometime random characters in certain parts of command lines. I will continue to monitor the blocks and adjust the LLM instructions as we go, but they should be pretty close now. The command line instructions are brand new, so it is not a surprise that the need a little tweaking. Besides that, there was one other false positive for a Hasleo binary, so the other instructions are holding up nicely. Thank you guys!