[liberationtech] How to defend against attacks on chips?

[Steve Weis]

My company is working on the problem of how to compute on untrusted
platforms. We gave a technical talk earlier in the year about
privilege escalation through physical attacks:
http://cansecwest.com/slides/2013/PrivateCore%20CSW%202013.pdf

>From a practical perspective on x86 platforms, we can deal with
compromised boot integrity, malicious PCI devices, or non-volatile
memory, but still need to rely on a trusted CPU and TPM.

There are potential issues with the LPC bus and QPI. JTAG is an open
question on some platforms. Side channel attacks and fault injection
are also an open issue in general. I think the high-value target in
the ecosystem are microcode update signing keys. There was some
interesting fault injection research into Intel's microcode updates
recently: http://inertiawar.com/microcode/

>From a theoretical crypto perspective, there are protocols to safely
compute on malicious parties, although you'd still need some trusted
party to receive the results. Fully homomorphic encryption is an
example that could "efficiently" implement some secure computation
protocols. Efficient here means a trillion times slowdown -- although
that has been reduced a couple orders of magnitude in the last few
years. These crypto approaches also require redesigning or recompiling
most applications.

So, to answer your question: Today we need to trust at least the CPU.
Organizations that care about this tightly control their supply chain
or build their own hardware.

Here are some general links to physical attack references that my
company posted:
Physical memory attacks:
http://privatecore.com/resources-overview/physical-memory-attacks/
Trusted execution and server attestation:
http://privatecore.com/resources-overview/server-attestation/

On Sat, Jun 15, 2013 at 3:19 PM, Anthony Papillion
<anthony at cajuntechie.org> wrote:
> So we know the NSA is spying on the word. We know pretty much how they
> do it and we know that at least part of that spying and data collection
> is likely done by exploiting holes in software. We can fix that. We can
> move people to better software, not rely on software from companies who
> routinely turn over data, push open software, etc.
>
> But how do we handle hardware attacks? For example, what happens when a
> chip maker, say Intel, collaborates with the government to allow access
> to users systems from the chip level? How can we defend against this?
>