QUOTE (pragma @ Mar 8 2014, 03:33 AM) 
Thanks for the perspective. I believe the situation I'm concerned about is covered by your discussion above. Wearing my security designer hat, I want to put a camera on the outside of a building in order to spot potential intruders. I don't want to pay a tremendously expensive professional like a decker to protect it all the time, so I'd like to slave it to a host. (This gains the additional benefit that an on call decker can show up to defend the host from anywhere in the matrix.) However, if I do so, then anyone within 100 meters of the building can get marks on my host, and one mark is all it takes for my precious paydata to leak out the walls without anyone stepping into line of sight of the camera.
OK, so the fix is that I have a separate security host and paydata host. However, this still isn't perfect; my entire security can be compromised by anyone standing within 100 meters of my building. I can put up a fence 200 meters away, and pay lots of people and dogs to wander around at night. However, this makes building any facility with even a modicum of security and technial sophistication really, really expensive. Hosts just don't make sense in many situations; even in the heavily secured facility I described above I'd be really hesitant to use a setup which could lose entire zones of security at the same time. Given the choice I'd go with a network of on-site security spiders every time. That's a shame because I think having an asymmetry in hacking stationary, mildly hardened targets is cool and it really improves the verisimilitude of the world if there exists any way to secure data.
In summary, I agree with your assertion that "the security risk only exists for devices that are slaved to the host, but not physically secured," but I think the standard for physically secured is ridiculously high. Physically secured in this context means there is no way for an unauthorized human to stand within 100m of a slaved device at any time. Because of that, and the inability to localize failures, I have difficulty seeing why anyone would ever install a host.
Also, the rest of your statements raise a few questions for me.
OK, so the fix is that I have a separate security host and paydata host. However, this still isn't perfect; my entire security can be compromised by anyone standing within 100 meters of my building. I can put up a fence 200 meters away, and pay lots of people and dogs to wander around at night. However, this makes building any facility with even a modicum of security and technial sophistication really, really expensive. Hosts just don't make sense in many situations; even in the heavily secured facility I described above I'd be really hesitant to use a setup which could lose entire zones of security at the same time. Given the choice I'd go with a network of on-site security spiders every time. That's a shame because I think having an asymmetry in hacking stationary, mildly hardened targets is cool and it really improves the verisimilitude of the world if there exists any way to secure data.
In summary, I agree with your assertion that "the security risk only exists for devices that are slaved to the host, but not physically secured," but I think the standard for physically secured is ridiculously high. Physically secured in this context means there is no way for an unauthorized human to stand within 100m of a slaved device at any time. Because of that, and the inability to localize failures, I have difficulty seeing why anyone would ever install a host.
Also, the rest of your statements raise a few questions for me.
- * What's the difference between being attached to a host and slaved to a host? The only network relationships described in the book (to my knowledge) are master/slave relationships.
* I think the errata changed the wording of hosts and IC such that IC share marks on you, but you have to mark each piece of IC individually.
100 meters is not direct connection. direct connection is you, sitting there, with a cable running from your cyberdeck directly to the camera.
from 100 meters away, you can detect the camera and hack it, but it still enjoys all the benefits of being slaved to the host, which typically means that it would be just as easy to find the host itself and hack into it. the only benefit to hacking the camera from 100 meters away is that if you manage to beat the host's dice pools, you will have a mark on both the camera and the host, whereas if you beat the host's dicepools while hacking the host itself you will only have a mark on the host. of course, once you've gotten into the host, getting a mark on the camera as well generally becomes pretty easy, but you still had to go up against the host's defences and win either way.
edit: oh, and as far as the errata... that is incredibly confusing. before the errata, i knew exactly what it did. now, i read that and the sentence looks incomplete... i think it needs to actually specify what is being marked.