i seem to be getting a lot of mileage out of the NSS folks lately. one might think it's because i have some sort of grudge or maybe that i'm one of the av industry's 'pets' and just defending my master (like i did back when i accused a big chuck of the av industry of being snake oil peddlars, i'm sure). the truth is, however, that NSS Labs have simply given me a wealth of material to work with. please rick or vikram, talk some more.
in a previous post i linked to this video taken at the source boston conference this year. it's a panel discussion with vikram phatak, peter stelzhammer, and mario vuksan, with andrew jaquith acting as moderator. i'm going to be referring to it again in this post - specifically at about 25:00 minutes into the video where vikram talks about exploits.
now i'm going to be generous and overlook the fact that he confuses exploits with vulnerabilities at one point. it's vulnerabilities that are the flaws, exploits are what takes advantage of the flaws. the truth is public speaking isn't necessarily easy and things can come out wrong (or not come out at all), as i found out not too long ago, so it seems reasonable enough to me that he actually meant vulnerability when he said "it's an exploit, it's a flaw in IE or in Firefox or..."
what doesn't seem reasonable to me, however, is the idea he was actually talking about vulnerabilities rather than exploits for that entire part of the discussion. i'm fairly certain he was talking about exploits for the most part, so when he trotted out this idea that there are no samples when it comes to exploits my immediate reaction was "that's bullshit".
we exist in a universe of cause and effect. well, ok, determinism breaks down at the quantum level but at the macroscopic level determinism is the overriding trend - and even more so when we're talking about digital computers because they are deterministic finite state machines. the exploitation of a vulnerability doesn't just happen by magic, this isn't some sort of cyber-voodoo, there is a causal agent, an actor, some chunk of data that the computer receives which, when the computer operates on it, results in the vulnerability being exploited.
and guess what that chunk of data is - that's right, it's the supposedly fictional exploit sample. it doesn't matter that it's not an *.exe or that it may not even be saved to disk, that just makes collecting it more challenging. when asked for a sample, that chunk of data (along with a way to replay it's delivery to the vulnerable system/subsystem) is what is being requested - an example of the agent which causes the vulnerability to be exploited.
at this point some of you might be thinking about the fact that an exploit for a particular vulnerability can take many different forms and might even be thinking that that would make the previously described sample worthless. 'many forms' is not a new concept in the anti-malware field. the fact that an exploit can come in many forms is just another aspect of polymorphism. that's probably not a word that gets associated with exploits very much, in part because of a rather arbitrary convention of differentiating between 'code' and data. code is data, however, and any data the computer bases a decision on is for all intents and purposes the same as an instruction (the data dictates what the computer will do next) and can be considered a kind of code. so while exploits may not superficially resemble other polymorphic things from the past like polymorphic viruses or server-side polymorphic trojans, exploits are programs of a sort and multiple different exploits for the same vulnerability are functionally equivalent programs - which, when considered in that light, actually does bear some conceptual similarity to server-side polymorphic trojans. as such an exploit sample doesn't become worthless just because there are other forms it could take, it just means that detecting that one form represented by the sample is only the beginning.