Yeah, Yeah dictionary attacks... The key is that the search space is actually thinly populated enough to make dictionary attack hard. Most usernames are 6 characters or more, many include numbers, that is about 26^6 worth of search space per domain. Of course this is not evenly populated, but the odd thing is that the usernames turn out to be more random than the average password. This is because random is not unguessable. Many usernames are surnames, many are compounds of initial plus surname, only a relative handfull are commonly used names and those tend to get grabbed fast. so you have a pretty big search space, millions of possibilities and that for each one of fifty million domains. The same does not hold for do-not-call lists. The problem there is that something like 80% of the numbers available at active exchanges are already allocated. Most of the stock of unused numbers are on exchanges that have not yet been allocated. Since something like 30% of subscribers sign up for do not call the result is that dictonary attacks are easy. Also we add out of service addresses that get spammed anyway to the list. So the list is not an accurate way to find out if an address is in service or not. Alan knows quite a few addresses that get spammed that are invalid.

-----Original Message----- From: Hallam-Baker, Phillip [mailto:pbaker@verisign.com] Sent: Friday, November 21, 2003 7:21 PM To: 'Steve Schear' Cc: cypherpunks@lne.com; asrg@ietf.org Subject: RE: [Asrg] Re: [Politech] Congress finally poised to vote on anti -spam bill [sp]

We need to consider the technical workings of the do-not-spam list and the requirements that we would like the FTC to meet.

I propose as a minimum:

1) Allow individual subscribers to list their email addresses with the service. 2) Permit mail sender to quickly determine whether a given email is on the list 3) Be distributable in a form that does not permit use as a mailing list. 4) Permit the storage of attributes in association with each listing, minimally the date of subscription.

In addition we might add:

5) Allow domain name owners to list their domains. 6) Provide for authentication of listing requests

These requirements can be met using completely generic and to my knowledge unencumbered technology. For the purposes of avoiding patent encumberabces I disclose the following - I published note on the basic idea of using a one way hash to conceal an email address on a do not spam list in 1995, I also implemented the scheme at that time. The idea is not entirely novel, hash databases have been used for at least twenty years, there may also be similar ideas in the cryptography litterature.

My proposal would be to use a message authentication function such as HMAC-SHA1 with a key such as SHA1 ("FTC Do Not Spam List") to create a unique digest function for the purpose. There is a security consideration here, use of a digest such as SHA1(email) might lead to chosen protocol attacks.

To add an individual email address "alice@example.com" to the list we calculate HMAC ("alice@example.com") to create the key. A domain may be represented by the string "example.com".

To determine whether the address "bob@example.com" is on the list it is necessary to test for both the specific email address and the domain.

[This can be made to meet arbitrarily complex requirements]

The list is distributed as a set of key/value pairs. Sorting the list according to the key values allows rapid lookups by means of binary search, or since the hash function is guaranteed homogenous using ranged search using the hash value as an estimator for the index position. It is not necessary to distribute the list sorted.

There are also a few tricks that can be used to reduce the usefulness of such a list for address validation.

This same concept can be used to conceal the filter terms used in cersorware.

Phill

_______________________________________________ Asrg mailing list Asrg@ietf.org https://www1.ietf.org/mailman/listinfo/asrg