OmniDefender - New Antivirus Software 2025

There have been improvements, but it's still just as heavy.
I didn't play with any recent malware, I only used 2 VMs with one attacking PC and one victim PC.

For a start, it's still as heavy as ever....

Capture d’écran 2025-07-30 012019.png

I'm testing a RAT (Xworm): detected

Capture d’écran 2025-07-30 013301.png

Test of a basic Async, also detected.

Capture d’écran 2025-07-30 013840.png

I'll complicate things a bit: I'll steal the digital signature of a program (Adobe Reader) and I'll simply encrypt the RAT's Stub and create an Autostart entry (named coucou.exe ).
Omni doesn't block it, my VM1 connects to my VM2

Capture d’écran 2025-07-30 014048.png
Capture d’écran 2025-07-30 014251.png
 
Welcome to MT! This is the French product that I mentioned on MT recently that I discovered while researching and working on LLM analysis as well. LLM malware analysis is the future, so you are on the right track! If you do not mind me asking, did you decide to ditch the local LLM analysis? I could not get local models to work reliably and to return the correct verdict consistently (there were simply not enough parameters), so we moved to the cloud. But I bet in the next year or so there will be local models that will work great, maybe even with CPU only analysis. Anyway, great job, keep going!
 
There have been improvements, but it's still just as heavy.
I didn't play with any recent malware, I only used 2 VMs with one attacking PC and one victim PC.

For a start, it's still as heavy as ever....


I'm testing a RAT (Xworm): detected


Test of a basic Async, also detected.


I'll complicate things a bit: I'll steal the digital signature of a program (Adobe Reader) and I'll simply encrypt the RAT's Stub and create an Autostart entry (named coucou.exe ).
Omni doesn't block it, my VM1 connects to my VM2

It seems that the developer has implemented LSH (most likely Trend Micro Locality Sensitive Hash). That’s a good initiative. But again, it depends on the volume of threat intelligence as well as whether or not machine learning is used on these hashes so they can be clustered.
 
Virtual Citizens, i have a proof that this OsirisXD is a pure SCAMMER, don't put any credential neither your creditcard i warn you

date 2025-07-31 eu time 003847.png
.

i made a test to see if i reached a real dns, and nothing !

i made a test to see if i reached a FREE download and i went through an inscription process, wich is not FREE anymore

i made a test to see if i have a secure cookie to have some information from the server and i just had a 304 mean nothing is modified just to pass on a 200 (statut ok)
but the cookie is not secure and i have only 2 cookie so it's mean the plugin from google, microsoft and apple you see is not really conneccted to there own server.

i made a test to see if a normal inscription is allowed and no you don't have any warning message, even my cookies are blocked :


Normaly this line above is secure and you can not see what is behind it.
the thing is there is even no HSTS on his web storage from firebase servers (no certificate authority)

this is unbeleivable that someone who pretend to be a security researcher, has been marked as a good guy by all te security expert here, without making some research too.

have a nice day ^^
 
Last edited:
OmniDefender Version 1.2 has been released.
New Features: The kernel now analyzes the file invoked by interpreters which includes but is not limited to .vbs, .vbe, .js, .cmd, .hta, .msi, .py, .pyc .r, pre-execution. They are blocked before being given the chance to run.
Fixed the Process Tree in Real Time Protection causing instabilities
Added more Sanity Checks around Real Time Protection and Smart Scan



Issues have been found with updating the database and will be patched in version 1.3
 
Welcome to MT! This is the French product that I mentioned on MT recently that I discovered while researching and working on LLM analysis as well. LLM malware analysis is the future, so you are on the right track! If you do not mind me asking, did you decide to ditch the local LLM analysis? I could not get local models to work reliably and to return the correct verdict consistently (there were simply not enough parameters), so we moved to the cloud. But I bet in the next year or so there will be local models that will work great, maybe even with CPU only analysis. Anyway, great job, keep going!
Sorry for the late reply, I appreciate the warm welcome!

Yes I ditched local LLM for now as inference was much slower and can be resource intensive over extended periods which would affect user experience. But it's not completely out of the table. As you stated, local LLMs are becoming more efficient and less resource intensive.
 
  • +Reputation
Reactions: Trident
View attachment 289979
This happens because your real time protection still broken. After I do restart and run scan most of active malware gone but very destructive ones left then my vm got destroyed.
This isn't a Real Time Protection issue at all. Most of them are suspended and blocked from the CPU Tab, thats how OmiDefender's kernel detects and blocks malware, the suspended malware never actually execute. But seeing them in process explorer without context does give a false impression until you look at the CPU Tab. This isn't a Real Time Protection issue nor a database issue but the Kernel Driver logic which blocks at process creation. In future versions we'll work on detecting malware at an earlier stage so they don't appear in process explorer at all.
 
Yeah because it suspends all malwares but not deleting them.
Yes, with pro-active detection, executables tend to open file handles at process creation so attempting to delete them at that stage results in "This process is already open in (filename)" so only deletion and reboot is possible. That's one of the reasons why many antivirus ask to restart the PC when malware is detected as It allows for efficient cleanup without any obstacles. As mentioned earlier, this is a kernel driver logic, that will be updated.