Re: Extracting signature snippets from AV databases

Yuri Slobodyanyuk wrote:

> SideNote: few years ago I watched the heated dabate on some forum (don't
> remember any details) where AV vendor representative was accusing
> open-source AV developers of reverse-engineering the virus-signatures
> instead of gathering their own, so logic
> says it has been done before by someone.

Yes -- the Open AntiVirus group had a "signature extractor" that 
basically took a sample of a piece of malware detected by a scanner 
then successively munged it (overwriting various sized and location 
blocks with nulls IIRC) until the scanner didn't detect it.  Applying 
this approach from several starting points and iterating eventually 
gives you a suitably small-ish  "chunk" of the original file that 
appears necessary to its detection, at least relative to the specific 
scanner in the harness.  Said "chunk" was then added to OAV's detection 
database.

For a dumb, brute-force string scanner like OAV's and for some simple 
types of malware this can produce marginally useful "signatures", if 
detection of relatively static objects (such as non-morphing malware, 
which includes most self-mailers) is your objective.

It is probably even a defensible business model if you have no ethics.

However, taking such a "signature" and sticking it into an arbitrary 
file at an arbitrary offset (as the OP is apparently planning on doing) 
is not even guaranteed to trigger the original scanner such a 
"signature" was extracted from, for reasons I mentioned in my earlier 
post and also described by Robert Sandilands.

That the OP was apparently unaware of these basic issues and 
limitations of his proposed approach is rather worrying, given he is 
the developer of a security product.


-- 
Nick FitzGerald
Computer Virus Consulting Ltd.
Ph/FAX: +64 3 3267092