Challenge to David Wagner on TCPA

[AARG!Anonymous]

Brian LaMacchia writes:

> So the complexity isn't in how the keys get initialized on the SCP (hey, it
> could be some crazy little hobbit named Mel who runs around to every machine
> and puts them in with a magic wand).  The complexity is in the keying
> infrastructure and the set of signed statements (certificates, for lack of a
> better word) that convey information about how the keys were generated &
> stored.  Those statements need to be able to represent to other applications
> what protocols were followed and precautions taken to protect the private
> key.  Assuming that there's something like a cert chain here, the root of
> this chain chould be an OEM, an IHV, a user, a federal agency, your company,
> etc. Whatever that root is, the application that's going to divulge secrets
> to the SCP needs to be convinced that the key can be trusted (in the
> security sense) not to divulge data encrypted to it to third parties.
> Palladium needs to look at the hardware certificates and reliably tell
> (under user control) what they are. Anyone can decide if they trust the
> system based on the information given; Palladium simply guarantees that it
> won't tell anyone your secrets without your explicit request..

This makes a lot of sense, especially for "closed" systems like business
LANs and WANs where there is a reasonable centralized authority who can
validate the security of the SCP keys.  I suggested some time back that
since most large businesses receive and configure their computers in
the IT department before making them available to employees, that would
be a time that they could issue private certs on the embedded SCP keys.
The employees' computers could then be configured to use these private
certs for their business computing.

However the larger vision of trusted computing leverages the global
internet and turns it into what is potentially a giant distributed
computer.  For this to work, for total strangers on the net to have
trust in the integrity of applications on each others' machines, will
require some kind of centralized trust infrastructure.  It may possibly
be multi-rooted but you will probably not be able to get away from
this requirement.

The main problem, it seems to me, is that validating the integrity of
the SCP keys cannot be done remotely.  You really need physical access
to the SCP to be able to know what key is inside it.  And even that
is not enough, if it is possible that the private key may also exist
outside, perhaps because the SCP was initialized by loading an externally
generated public/private key pair.  You not only need physical access,
you have to be there when the SCP is initialized.

In practice it seems that only the SCP manufacturer, or at best the OEM
who (re) initializes the SCP before installing it on the motherboard,
will be in a position to issue certificates.  No other central authorities
will have physical access to the chips on a near-universal scale at the
time of their creation and installation, which is necessary to allow
them to issue meaningful certs.  At least with the PGP "web of trust"
people could in principle validate their keys over the phone, and even
then most PGP users never got anyone to sign their keys.  An effective
web of trust seems much more difficult to achieve with Palladium, except
possibly in small groups that already trust each other anyway.

If we do end up with only a few trusted root keys, most internet-scale
trusted computing software is going to have those roots built in.
Those keys will be extremely valuable, potentially even more so than
Verisign's root keys, because trusted computing is actually a far more
powerful technology than the trivial things done today with PKI.  I hope
the Palladium designers give serious thought to the issue of how those
trusted root keys can be protected appropriately.  It's not going to be
enough to say "it's not our problem".  For trusted computing to reach
its potential, security has to be engineered into the system from the
beginning - and that security must start at the root!

---------------------------------------------------------------------
The Cryptography Mailing List
Unsubscribe by sending "unsubscribe cryptography" to majordomo at wasabisystems.com