[tor-talk] Intel ME / AMT + NSL vs Tor Nodes

podmo podmo at sigaint.org
Sat Dec 17 21:48:51 UTC 2016

> https://www.reddit.com/r/onions/comments/5i6qa3/can_the_nsafbi_use_intel_me_to_defeat_tor_on_95/
> "
> So, The NSA and FBI CAN force Intel to give up the keys Intel ME.

There is a well-written, rational response further down in that same
Reddit thread:

"This post has a lot of misunderstandings behind it. First off, the
Intelligence Community does not need to force Intel to give up
Manageability Engine keys (or AMD's PSP keys for that matter). Both the
keys and the toolchain, as well as the source code are traded underground.
I know that at least up to firmware version 8 is traded underground, and
version 11 (the latest) is available without difficulty to people who know
how to find it. I have access to version 8's signing keys myself, being in
that scene, but all my computers use version 11 so I haven't cared to mess
with it. It's certainly not common but it is absolutely something that
FVEY and related contractors (Raytheon, Leidos, half the people you'll see
at ISS, etc) will be able to get their hands on, if they haven't already.

Second, the abilities of the Manageability Engine are greatly
over-exaggerated. It cannot be used to access all your data remotely. That
only works if you have all AMT features enabled, and you have a special
device called a BMC card plugged into your computer and connected to the
network. BMC cards can include 3G/4G or WiMax support, which is where the
myth that vPro CPUs have a 3G backdoor comes from. I have an enterprise
ThinkPad that proudly boasts having WiMax support, requiring extensive
configuration. It was expensive. If you don't have a BMC card (and you do
not), then it is not possible to remotely control your system. Even if you
did have a BMC, simply having the signing keys and toolchain for the ME
would not be sufficient to get in. An attacker would need either a 0day,
or your credentials. Having the signing key allows nothing more than
writing malicious firmware over SPI and allowing it to persist. It's just
a little more powerful than the UEFI kits cr4sh can write, and just as
easily detectable by reading your flash chip. But it's not like you're
analyzing your microcode (of which there are likely signing keys being
traded as well), which can also be installed on a large number of systems,
considering the BIOS functions to load the latest microcode it has into
the CPU.

Thirdly, you don't have to worry about the ME hiding Intel-provided
backdoors because it is not impossible to reverse engineer ME firmware.
The firmware is huffman coded, which can be decoded with some manual
effort, and then you have ARCompact bytecode with Java-based modules.
Intel can be a nasty company, but they aren't going to risk everything
with overt backdoors that simply exfiltrate your memory over the network.
Plus you could easily block that with a separate firewall. Even if it is
sent out-of-band with regards to the kernel's networking stack, it's still
sent over the same physical NIC, just with a different IP and MAC.

The ME is absolutely not what you have to worry about in these threat
models. It is only a way for malware to hide itself from forensic
analysis, not a mystical way to remotely contact any system which runs it,
absent a BMC card.

If you have to have something to worry about, worry about 0days. They are
much more dangerous and valuable than something which, at best, provides a
persistent infection that is trivial to detect offline. There are RCEs for
every major httpd. There are LPEs that even work on grsecurity (at least
one that I know of), and dozens that work on vanilla Linux. There are at
least two traded ring 0 RCEs for Windows, one of which I have, and there
are probably a couple ring 0 RCEs in Linux's Netfilter (conntrack,
anyone?). Secure your OS, use sandboxes and mandatory access controls
(SELinux or AppArmor or RBAC), keep up to date, read security mailing
lists, be wary of red herrings, use grsecurity + PaX, and most
importantly, understand your own threat model.

I can say with absolute confidence that the Intel Manageability engine is
not a threat in the least to the integrity of the Tor network. Especially
not when each and every one of you are running a browser which can be
exploited with images and CSS. Sandbox your shit."

P.S. Please double-check the facts before spreading FUD.

More information about the tor-talk mailing list