signing a robot's key - was: Re: Global Directory signatures

Jeff Fisher jeff+gnupg at
Sat Jan 1 22:40:27 CET 2005

Hash: SHA256

On Sat, Jan 01, 2005 at 08:33:56PM +0000, Neil Williams wrote:
> On Saturday 01 January 2005 6:00 pm, Jeff Fisher wrote:
> > For me, in this case, the key and the role are the same thing.
> That's the problem, right there. A signature is not for YOUR benefit! You need 
> to take on board how signatures are used by others. To others, the key and 
> the role, the key and the person, are completely separate.

In this case, there is no person.  There is a corporate entity, a keyserver,
and a web server.  No person.  The key exists for the role.  The trust in the
role would be very limited without the key. _In this case_, the key and the
role, even though they are not the same thing, are inseperable.  Take away
one, and the other has little reason to exist.   I'm not saying that is the
general case, I'm just stating my opinion of this particular case.

For the general role based case, my opinion is that some these keys can be
verified to a level that I would be confident enough to publicly sign them.

> The capacity is NOT in the system to say that your signatures should always be 
> untrusted - basically because the system is of the opinion that untrustworthy 
> signatures shouldn't be made in the first place!
> :-))

Ah, but the capacity is in the system to do just that.  Edit the trust of a
key, and there is an option 'I do NOT trust'.  I would think there is
something similar in the PGP code.  I haven't looked at the code, but I would
hope this setting overrides any trust earned from signatures on a key.

Indeed, the default is to not trust signatures unless you have a close link in
the WoT.  Trust is earned, based on the rules and signatures that _you_ make

> > However, if it is used for business / transactions, then I will not trust
> > anything but personally verifying the identity of the key, or a trusted
> > third party doing the same (but not through key signing).  For the ssl
> Leave x.509 out of this - it doesn't apply to the issues.

I'm saying that I don't trust signatures at all in this case. From anybody. In
this case. X.509 or PGP is irrelevant to the issue.

> The confusion is about keysigning of individual keys that are used by 
> machines. Never sign them!

The set 'keys used by machines', and 'role-based keys' are not neccesarily

My opinion is that people should be able to make the choice of what keys to
sign (publicly or privately) individually.  Fortunately, barring massive
changes to the OpenPGP spec and current implementations, this is the case.

> > example in another e-mail, this is getting the URL for the bank from their
> > literature, or a similar method.  If they used pgp, I would get the
> > fingerprint from their brochure.
> ??? I hope you are not saying you'd make an exportable signature on their key 
> on just that basis???

Ok, if you walk into your local branch of the bank, and they have a poster
with their PGP fingerprint, how much more authoritative do you need?  If
you're trusting the signature with your money, aren't you already saying to
the world that you have confidence that this key is the bank's key?  And no, I
would not make an exportable signature based on that, as I would expect people
to verify the key themselves when it's that important.

> > , or in the case of Robot CA's, to mark 
> > them as trusted introducers.  My setup probably needs some tuning to verify
> > this is working correctly, but as stated above, I won't trust anything
> > sensitive to pgp unless I really trust the person, and have personally
> > verified the key.
> Fine - so why are we discussing this at all??

Because I believe that someone could trust a role-based key enough to
broadcast that fact to the world.  That does not seem to be the commonly held
opinion on this list.

> The WoT is not a take-it-or-leave-it all or nothing construct. You retrieve 
> those keys that you know belong to people you already trust. You set your 
> personal trust level to tell GnuPG how much you trust that person to verify 
> someone else's key and ALL other keys, if you don't know the person, tell 
> GnuPG "don't know". There's an option in the trust that explicitly uses 
> "don't know". It's not a cop-out, it is a vitally important setting.

If my signing policy can break your WoT, then there is something wrong with it
at the conceptual level.  I don't think this is the case.  Based on the
difference in opinions, I don't think I would ever have enough signatures
close enough to your own to make my key in any way trusted on your keyring.
In this way, the WoT is working perfectly.  Without any work on your end, you
won't trust any of my signatures, and they won't enter into the trust
calculations on your keyring.  Great, ain't it?

- -- 
Me - jeff at


More information about the Gnupg-users mailing list