Serious Discussion Quick Play with McAfee

I am still working out the McAfee detections and logs, the log contains something like this (from another thread).

Detection SourceFile ReputationHTI Reputation
hti44 <- online reputation reports the file as malicious
cache00 <- nothing in cache, as detections are cached only once they occur, not before that
uwp00 <- not a UWP app
signature050 <- picked up by a Yara rule
trust-dat44 <- file is untrusted
rp-s44 <- minor tweaks to the final score. RealProtect static analysis deems the file suspicious
av01 <- these verdicts are minor tweaks to the final score
neo01 <- minor tweaks to the final score

The “Cache” appears to come into play when behavioural-based detections occur. McAfee likely refers to the behavioural blocking database (the one where actions are recorded) as “Cache”.

Detections from the “Cache” are named “Cache!<part of SHA256>
Any update on this ?
 
Like a guide. For example, if a file has been detected and you open the log to check it, using the guide you can break down and see the overall picture of the detection 🙃
There is no really guide that can be created, there are many engines and every engine has offline and online verdict. You will probably need to read this 3-4 times.

All the engines are linked to a fusion framework. This fusion framework although patented decision matrix, follows widely adopted understandings:
Every detection that is machine validated for false positives -> strong classifier.
Every detection that cannot be so thoroughly tested (this is ML/AI) -> weak classifier.

So all engines are executed on all files. Each one of the engines returns verdict. Several engines take into account the reputation of the certificates/signatures (all of them) and the reputation of the URL when the file is downloaded.
This alters the verdict they produce.

You then need to look in the json where it says final detection source. This is the engine whose verdict McAfee considered more reliable.

The engines are:
HTI (heuristic threat intelligence). This is the Artemis/JTI that uses mixture of reputation and heuristic. E.g low reputation file in user space used to be 196612 heuristic. The boolean conditions of this heuristic will not be met if the file was in C:/Windows or it had favourable rating. Obviously it’s not one rule, the list of rules used to be readable in the legacy version and there were hundreds of rules.

The most confusing part is HTI is listed as separate entity, but then every engine in this decision matrix has offline and HTI column. Likely, after every engine findings, HTI goes and looks for any rules matched. Without these rules confidence is x, if there is a match m, confidence is X + now the new value of Y.
Example: real protect static outputs confidence of 4. HTI contains rules that look for executables with low reputation, suspiciously modified timestamps, located in user space. The HTI value of the Real Protect Static engine is now higher.

Trust dat: this is local repository of rules and other data that allow McAfee to determine the trust level of a file when the cloud doesn’t have enough information.

Then you’ve got the quick winners:
AV: these are highly optimised generic detections that target very prevalent malware families significant enough to get profiles + generic tactics and techniques.

Neo: this is your in-memory sandbox emulator based on heuristics that also look for generic malicious behaviour. Based on the matched indicators, neo outputs different score.

Signature: McAfee no longer does signatures, they do Yara rules. Yara rules have quality level based on the noise they generate. This affects the final score (0-50).

Rp-s(tatic): this is the Real Protect static analysis, which is local but when the local model is in doubt, it consults the cloud one. This has been the case since Real Protect was introduced.

Then we come to the behavioural detections:
RP-fileless: this is the model that deals with fileless malware.
Cache: this is the database that McAfee uses to write the process actions
RP-d(ynamic): this is the core behavioural blocking.

It’s a very complex decision matrix and the JSON format doesn’t help.

I suggest you paste the log on AI and ask it to build a nice table.
 
Last edited:
I reinstalled McAfee on fresh Windows a few days ago. I think I just encountered the AI-generated threat description for the first time.

Screenshot 2025-10-19 221418.jpg
 
I reinstalled McAfee on fresh Windows a few days ago. I think I just encountered the AI-generated threat description for the first time.

I have to admit I am not hating that, it is quite unique.
Mitre ATT&CK for beginners.
 
@Sunqfu thanks for posting, I just did a manual update and now have dark mode. It adjusts to your Windows theme, as I didn't see an additional setting for it. And since McAfee has minimal settings, that falls in line with its "set it and forget it", approach.
The update needed the app restart, but didn't need a Windows restart.

update2.jpg

mcafee update.jpg
 
@Sunqfu thanks for posting, I just did a manual update and now have dark mode. It adjusts to your Windows theme, as I didn't see an additional setting for it. And since McAfee has minimal settings, that falls in line with its "set it and forget it", approach.
The update needed the app restart, but didn't need a Windows restart.

View attachment 292731
View attachment 292732
Both McAfee and I in my projects use Web View hahah

The launch speed of mine is a bit more optimised, I spent over 12 hours optimising it.

McAfee needs to recompile their UI to .net 10 which eventually will happen soon.
 
@Sunqfu thanks for posting, I just did a manual update and now have dark mode. It adjusts to your Windows theme, as I didn't see an additional setting for it. And since McAfee has minimal settings, that falls in line with its "set it and forget it", approach.
The update needed the app restart, but didn't need a Windows restart.

View attachment 292731
View attachment 292732
Is the white line a reminder that it was once a white app 🤣