I figured out at ridiculously simple but extremely effective method to fix the Command Line false positives. Earlier today I added 9 new LLM instructions, and so far, so good

. I initially thought that the command line analysis would be extremely similar to script analysis, and that we could just copy what we did from the script analysis to the command line analysis and we would achieve efficacies approaching 100. It turns out that they are quite different, and now it makes total sense. That is one of the cool things about all of this, you learn new stuff every day.
I understand that with any kind of AI, there is some skepticism. But I promise, Sirius is for real. I am baffled by why anyone would accept the old binary classification algos that all of the other "AI" cybersecurity products use (and the old VoodooAi as well), but do not seem to understand that we are doing basically the same thing with Sirius, but with a much, much, much smarter algo. Sure, it was extremely difficult to get right, but once it was right, I knew it was right.
We have the text / script attacks instructions nailed completely. The portable executable verdicts are almost spot on, and it will be super simple to figure out what metadata to add to detect the misses, as opposed to the shotgun / kitchen sink method that results in false positives (like all of the sandboxes). As mentioned above, the command line analysis is looking great, but there will be a few more instructions we need to add.
Anyway, still waiting on
@Shadowra to test SiriusGPT

.