Re: automatic signature generation

On 5/22/07, Tim <tim-security@xxxxxxxxxxxxxxxxxxx> wrote:
> Therefore, the early thought that comes
> into my mind is "creating an automated signature generation tool is as
> difficult as creating an automated attack generation tool". I would
> like to know your opinion on this.

I would say no. That is, I think it would be easier to create an
automated signature generation tool that it would be to create an
automated exploit generation tool. This is based on my experience with
machine learning algorithms and penetration testing. This of course
with the caveats:

- To create a signature for a single vulnerability, the generation tool
would need to have a set of exploits for that vulnerability and a
large body of harmless traffic to compare it against.
this is what I have in mind to start with. but there are problems. i
have manually created signatures for many vulnerabilities and for
various exploits/attacks, I had to use regexp or checks many fields
related to vulnerable protocols/applications. so we miss the
contaxt/semantics of the attack, if we directly apply machine
learning, at least to my understanding. if you know some work in this
direction, please refer. I would like to explore.

- The signature generation tool would not be able to generate
false-positive and false-negative free signatures (who does?).
However, for simpler cases the error rates could be quite low and
possibly even measurable.

As far as your comments about detecting flooding attacks, I think this
may actually be harder to get right.
under most general scenario, flooding is deected by the rate of
packets. so, if we keep checking the health of the victim
(destination), we can fine tune the threshold for this rate
automatically. you may like to see the work of J. Cannady on "CMAC and
flooding attacks"



Postdoc, DIT, University of Trento, Italy

Test Your IDS

Is your IDS deployed correctly?
Find out quickly and easily by testing it with real-world attacks from CORE IMPACT.
Go to to learn more.