Perry E. Metzger wrote:
> I think Steve is completely correct in the case of
> cryptography. We have a lot of experience of real
> world security failures these days, and they're not
> generally the sort that crypto would fix.

They are the sort that a different sort of way of using
crypto could fix.

>> Authentication is exactly what I need in the case of
>> spam/phishing:

Perry E. Metzger wrote:
> People have said that for quite some time. However, I
> doubt it would actually help. In the case of spam, all
> that would end up happening is vast amounts of CPU
> time being spent demonstrating that the made up
> addresses on spam were associated with actual RSA
> keys. (There is no practical limit to the number of
> RSA keys that may be generated.)

First, the phishing case:

Assume that instead of logging in through a possibly
hostile web page, people login using SRP built into the
browser chrome.  Then phishing goes away, because the
man in the middle gets no shared secrets.

Next, the spam case, including spam offering high yield
investments, spam promising millions of dollars stolen
from starving Africans, Indian made viagra, porn sites,
and two hundred and seventy tons of sugar.

To fix spam, we need automatic whitelisting plus
aggressive Baysian filtering.  At present this works
fine, no crypto needed, because the attacker seldom
bothers to adapt to your particular profile.  Tons of
spam descend upon me, and is magically banished, sight
unseen. When impersonation spam attacks become a serious
problem, then we will think about how to use
cryptography to beat them.  Sometimes I have to manually
whitelist people, which Grandma could not do, but that
is a user interface problem - observe that most IM
systems make whitelisting easy enough for Grandma.

> I would actually agree that we can implement operating
> system strategies that make malware harder to write. I
> don't know if it is likely that any current
> techniques, even including the nearly unheard of use
> of formal verification, would actually eliminate
> malware.

OLPC seems very nearly malware proof.  Malware would
require unusual privileges, and there is no easy way to
install software that requires unusual privileges on

The Cryptography Mailing List
Unsubscribe by sending "unsubscribe cryptography" to [EMAIL PROTECTED]

Reply via email to