Re:The Bogus CVE Problem
by cookiengineer on Friday September 22, 2023 @01:04PM
(#37614577)
I've followed the CVSS discussions kind of a lot, but I don't think it will fix the issues at hand. I think it would help a lot more if a score would reflect the stakeholders at hand for a vulnerability. Stakeholders in the sense of attacker, maintainer and user.
Maybe it could be represented with a score for each of the stakeholders, like for reporter/pentester, vendor/auditor, and customer? Example score could be: 4/8/6
The problem I also have with CVSS scores in general is that they don't reflect the _type_ of vulnerability. Adjacent vulnerabilities or RCEs should be scored separately because of their huge impact they can have on the network, compared to say, a local privilege escalation exploit that is useless without an exploit from another stage.
If a vulnerability allows quick lateral movements by the threat actors, they also increase likeliness of being used by them much more.
The combination of exploits in the sense of making an exploit feasible as an attack strategy also isn't reflected correctly, and that's what analysts on the defending side like to "abuse" to argue why they are too lazy to deal with it or why they won't update their systems. Usually the argumentation is something like "yeah but it's not feasible anyways", which is BS in the age of Exploit Service telegram channels that post a corporate zero-day, every day, for free.
The problems with all SIEM/SOAR systems I've seen is that they always reflect a point in time, and a decision that was made in that point in time. They don't represent an accumulative view of what actually is vulnerable, also what is _still vulnerable_ even when there was an alert about it a year ago. Log4j's vulnerabilities are the prime example again, too, because it proves that nobody has a correct asset inventory, and 78% of enterprises are vulnerable _again_ a year later because they reintroduced the vulnerable library without the Blueteam knowing about it.
Sorry for babbling too much about this, but I think that SIEM/SOAR systems don't work in practice and I've got a harsh opinion about them. Of course they can work for some situations, but it's still a luck shot if your analyst actually is an analyst or just a paper pusher hired directly from University.
We need to increase the quality of measurements, not the amount of measurements. And that's where all SIEM systems fail at. Petabytes of data per day are pretty much the definition of a needle in a haystack, and they don't help anyone if it gets more and more. And your magical AI unicorn won't fix this either, because of lack of reliable datasets (hence the discussion about CVE).
Maybe it could be represented with a score for each of the stakeholders, like for reporter/pentester, vendor/auditor, and customer? Example score could be: 4/8/6
The problem I also have with CVSS scores in general is that they don't reflect the _type_ of vulnerability. Adjacent vulnerabilities or RCEs should be scored separately because of their huge impact they can have on the network, compared to say, a local privilege escalation exploit that is useless without an exploit from another stage.
If a vulnerability allows quick lateral movements by the threat actors, they also increase likeliness of being used by them much more.
The combination of exploits in the sense of making an exploit feasible as an attack strategy also isn't reflected correctly, and that's what analysts on the defending side like to "abuse" to argue why they are too lazy to deal with it or why they won't update their systems. Usually the argumentation is something like "yeah but it's not feasible anyways", which is BS in the age of Exploit Service telegram channels that post a corporate zero-day, every day, for free.
The problems with all SIEM/SOAR systems I've seen is that they always reflect a point in time, and a decision that was made in that point in time. They don't represent an accumulative view of what actually is vulnerable, also what is _still vulnerable_ even when there was an alert about it a year ago. Log4j's vulnerabilities are the prime example again, too, because it proves that nobody has a correct asset inventory, and 78% of enterprises are vulnerable _again_ a year later because they reintroduced the vulnerable library without the Blueteam knowing about it.
Sorry for babbling too much about this, but I think that SIEM/SOAR systems don't work in practice and I've got a harsh opinion about them. Of course they can work for some situations, but it's still a luck shot if your analyst actually is an analyst or just a paper pusher hired directly from University.
We need to increase the quality of measurements, not the amount of measurements. And that's where all SIEM systems fail at. Petabytes of data per day are pretty much the definition of a needle in a haystack, and they don't help anyone if it gets more and more. And your magical AI unicorn won't fix this either, because of lack of reliable datasets (hence the discussion about CVE).