Re: Problems with SPF, solutions, and a timeline.

Peter J. Holzer wrote:
> On 2004-02-28 10:37:27 -0500, Alan DeKok wrote:
>>John Levine <[email protected]> wrote:
>>>I'd like to stop getting bounce messages, too, but it sounds like
>>>you're agreeing that LMAP schemes will not in fact deter any spam
>>>since spammers can easily evade it by using either their own throwaway
>>>domains or any of the millions of domains that will never publish LMAP
>>>records.
>>
>>  The same argument can be applied to any proposed anti-spam system:
>>There are tens of millions of systems which will never publish
>>anti-spam information "foo", or will never clean the spam trojans off
>>their systems, or will never fix their open proxy, or will never do
>>anything.
> 
> That's a different problem. Sure, there will always be such systems. The
> question is whether we get enough legitimate mail from such systems to
> bother. If we do, then the anti-spam system won't be much good - tough
> luck. But John's objection is on another level: If it doesn't stop spam
> even if it is fully deployed, because spammers can simply publish LMAP
> information like anybody else, then that scheme doesn't do what it is
> designed to do (it may have some other benefit, like stopping bounce
> messages, though).

Those bounce messages are spam, as are viruses with forged return addresses 
sent by integrated SMTP engines. Thus, viruses will be forced into using the 
domain of the user whose computer it infected, which is at least a little 
bit better. And if that user's SMTP server authenticates before sending, 
then the virus can be tracked to a specific user's computer. And if that 
SMTP server scans outbound messages for viruses, they will never hit the 
Internet at all.

>>  This sounds to me like an argument for throwing away SMTP entirely,
>>and starting over with perhaps a similar protocol on a different port,
>>but one which doesn't have SMTP's flaws.
> 
> Which might be a good idea. It will take some time to catch on, but if
> we look at the instant messaging or peer-to-peer protocols, users are
> quite fast at accepting new protocols if they think it is worthwhile.

Users aren't accepting protocols; users are accepting applications which use 
specific protocols on a specific network. Look, for contrast, at the success 
of xmpp: it may be hard to find, compared to, for example, AIM's, MSN's, and 
Yahoo's protocols.

>>>If so, why bother?  We should move directly to a signature scheme like
>>>domain keys in which each protected message makes a directly
>>>verifiable assertion that it's good,
>>
>>  And the spammers will (of course) sign their own messages, so such
>>schemes are pointless, because all they do is make the recipient do
>>more work, with no long-term benefit.
> 
> If you assume that signing alone will deter spammers, yes. Signing a
> message with a private key K says nothing more than that the message was
> sent by (or through) the holder of private key K. Of course a spammer
> can invent any number of private keys, so in itself that's useless.
> 
> To make a signing scheme work, you need a scheme to express "trust".
> Somebody has to certify that K is never used to send spam. This could be
> some CA, but it doesn't necessarily have to be. Something like the PGP
> web of trust or social networks like friendster or orkut would be
> preferrable IMHO.

This is not the only spam LMAP deals with. See above, please.

Philip Miller