Once you can define it you can code it and build it into the system that the wrong ones aren’t allowed to work at all.

Right. If it’s possible to detect a situation that is definitely an attack, then the software can simply prohibit it. The trickier problems have to do with things like SSL, where encryption is safer but (as you pointed out) it’s not currently practical to expect every website to use encryption.

And petnames, while a good idea, still need to rely on some form of central, unique identifier (like DNS) which can be attacked.

No, they don’t. Petnames are labels for unforgeable keys. No central coordination is necessary to use petnames.

“A better strategy might be to always show some type of indicator and have it look obviously right when things are right and obviously wrong when things are wrong.” I’d suggest that being able to define what is “obviously wrong” and what is “obviously right” (with nothing missed out in the middle) is far more important than how you display it to the user. Once you can define it you can code it and build it into the system that the wrong ones aren’t allowed to work at all.

And petnames, while a good idea, still need to rely on some form of central, unique identifier (like DNS) which can be attacked.

Once we have a petname-only world, we have a solution that satisfies your criteria of showing some type of indicator that is obviously right or wrong. Of course that depends on one’s definition of right and wrong. I’m assuming that a Petname is about as right as you can get, and the absence of a petname is about as wrong as you can get (at least, for humans).

In terms of your own work, it comes down to the principle of least surprise: “is the site I’m looking at the one *I* *named* in that label?” Petnames all but guarantee this to be case. DNS cannot, but people assume it can, and are frequently surprised when DNS does not meet expectations.

the only robust solution is to make encryption and authentication the norm as you suggest; any deviation from that would be considered an exceptional and rare event. this is Hard ™ - changing long familiar user behavior.

the processing cost of encryption is now close to negligable: VIA’s C5P core can process 1,832,151,000 bytes of AES ECB / sec (half that for CBC) for instance. [this is also less susceptible to side channel attacks like cache timing and power usage]

the expensive and most vulnerable step in the process will be key distribution for trusted parties - i’m not sure how identity management would be implemented for a process and large and varied as software development and distribution.

microsoft released something interesting recently:
http://www.identityblog.com/stories/2005/05/13/TheLawsOfIdentity.html
“The Laws of Identity”

the inclusion of “Directed Identity” is interesting and probably relevant.

Gregory Bateson pointed out long ago that Nothing can be a cause. So the non-arrival of a message may convey significant information. (As Bateson puts it, it is “a difference that makes a difference”.)