[Cryptography] On the deployment of client-side certs

[Philipp Gühring]

Hi,

> good commands from the bad commands.  That's the standard response to
> "we have
> a EAL 6.023x10^23 certified HSM!", to which you add "...that does
> absolutely
> everything the Windows PC it's plugged into tells it to".  So in the
> end you
> fall back to justifying the HSM as being useful for auditing purposes
> because
> an auditor can check off the physical artefact that The One Key is
> stored in.

Yes, I tried to argued that too for some time, that HSMs for
keymanagement/keysecurity are somewhat superfluous, since they are not
sufficient to protect a PKI.
The practical example that drove the point home was the DigiNotar hack,
where they demonstrably brought down not just the HSM's security model,
but the whole company, a good part of the government and thereby the whole
nation.
Because the attackers told the HSM to sign their keys, but since the HSM
only used the key but did not know or record what it was used for,
DigiNotar could not revoke the fraudulent certificates that were signed by
their FIPS certified HSM.

The light at the end of the tunnel was Apple recently who built an
ingenious security concept based on HSM's, which convinced me from the
conceptual point of view, and prooved that at least someone was able to
build a reasonable security system on HSMs, so we don't need to give up
the HSM form factor as a whole. 
And there Apple developed their own software on their HSMs.
So if you want a secure system, you have to build your own software on HSMs.
Just running standard PKCS#11 or similar software on it is not secure enough.
If you want to run a PKI, develop your certificate issueing software
inside the HSM.

> In practice it's a bit more complex than that, I looked at it in my
> 1998
> Usenix Security paper, see pages 4-5 of
> http://www.cypherpunks.to/~peter/usenix00_slides.pdf, specifically the
> "Tier 1
> ... 5" distinction.  Most crypto devices are Tier 1 or 2...

Best regards,
Philipp