From: Matt Blaze <mab@research.att.com>
Date: Wed, 2 Feb 94 13:11:01 PST
To: cypherpunks@toad.com
Subject: Notes on key escrow meeting with NSA
Message-ID: <9402022105.AA18514@big.l1135.att.com>
MIME-Version: 1.0
Content-Type: text/plain


A group from NSA and FBI met the other day with a group of us at Bell
Labs to discuss the key escrow proposal.  They were surprisingly
forthcoming and open to discussion and debate, and were willing to at
least listen to hard questions.  They didn't object when asked if we
could summarize what we learned to the net.  Incidentally, the people
at the meeting seemed to base a large part of their understanding of
public opinion on Usenet postings.  Postings to sci.crypt and
talk.politics.crypto seem to actually have an influence on our
government.

A number of things came out at the meeting that we didn't previously
know or that clarified previously released information.  What follows
is a rough summary; needless to say, nothing here should be taken as
gospel, or representing the official positions of anybody.  Also,
nothing here should be taken as an endorsement of key escrow, clipper,
or anything else by the authors; we're just reporting.  These notes
are based on the collective memory of Steve Bellovin, Matt Blaze, Jack
Lacy, and Mike Reiter; there may be errors or misunderstandings.
Please forgive the rough style.  Note also the use of "~ ~" for
'approximate quotes' (a marvelous Whit Diffie-ism).

NSA's stated goals and motives for all this:
	* DES is at the end of its useful life
	* Sensitive, unclassified government data needs protection
	* This should be made available to US Citizens
	* US business data abroad especially needs protection
	* The new technology should not preclude law enforcement access

They indicated that the thinking was not that criminals would use key
escrowed crypto, but that they should not field a system that
criminals could easily use against them.  The existence of key escrow
would deter them from using crypto in the first place.  The FBI
representative said that they expect to catch "~only the stupid
criminals~" through the escrow system.

Another stated reason for key escrow is that they do not think that
even government-spec crypto devices can be kept physically secure.
They do expect enough to be diverted to the black market that they feel
they need a response.  NSA's emphasis was on the foreign black market...

There seems to be a desire to manipulate the market, by having the
fixed cost of key escrow cryptography amortized over the government
market.  Any private sector devices would have to sell a much larger
number of units to compete on price.  (This was somewhere between an
implication and an explicit statement on their part.)

When asked about cryptography in software, "~...if you want US
government cryptography, you must do it with hardware~".

Clipper chips should be available (to product vendors) in June.  You
can't just buy loose chips - they have to be installed in approved
products.  Your application interface has to be approved by NIST for
you to get your hands on the chips.

An interesting point came up about the reverse-engineering resistance
of the chips: they are designed to resist reverse engineering the data
in the chip without destroying the chip.  It is not clear (from the
information presented at the meeting) whether the chips are equally
resistant to destructive reverse-engineering to learn the skipjack
algorithm.   They said the algorithm was patented, but they may have
been joking.  ("~And if that doesn't scare you enough, we'll turn the
patent over to PKP.~")

The resistance to reverse engineering is not considered absolute by
NSA.  They do feel that "~it would require the resources of a national
laboratory, and anyone with that much money can design their own
cryptosystem that's just as strong.~"

They repeated several times that there are "~no plans to regulate the
use of alternate encryption within the US by US citizens.~"  They also
indicated they "~weren't naive~" and didn't think that they could if
they wanted to.

There were 919 authorized wiretaps, and 10,000 pen register monitors,
in 1992.  They do not have any figures yet on how often cryptography
was used to frustrate wiretaps.

They do not yet have a production version of the "decoder" box used by
law enforcement.  Initially, the family key will be split (by the same
XOR method) and handled by two different people in the athorized
agencies.  There is presently only one family key.  The specifications
of the escrow exploitation mechanism are not yet final, either; they
are considering the possibility of having the central site strip off
the outer layers of encryption, and only sending the session key back
to the decoder box.

The escrow authorities will NOT require presentation of a court order
prior to releasing the keys.  Instead, the agency will fill out a form
certifying that they have a legal authorization.  This is also backed
up with a separate confirmation from the prosecutor's office.  The
escrow agencies will supply any key requested and will not themselves
verify that the keys requested are associated with the particular court
order.

The NSA did not answer a question as to whether the national security
community would obtain keys from the same escrow mechanism for their
(legally authorized) intelligence gathering or whether some other
mechanism would exist for them to get the keys.

The masks for the Clipper/Capstone chip are unclassified (but are
protected by trade secret) and the chips can be produced in an
unclassified foundry.  Part of the programming in the secure vault
includes "~installing part of the Skipjack algorithm.~" Later
discussion indicated that the part of the algorithm installed in the
secure vault are the "S-tables", suggesting that perhaps unprogrammed
Clipper chips can be programmed to implement other 80-bit key, 32 round
ciphers.

The Capstone chip includes an ARM-6 RISC processor that can be used for
other things when no cryptographic functions are performed.  In
particular, it can be used by vendors as their own on-board processor.
The I/O to the processor is shut off when a crypto operation is in
progress.

They passed around a Tessera PCMCIA (type 1) card.  These cards
contain a Capstone chip and can be used by general purpose PC
applications.  The cards themselves might not be export controlled.
(Unfortunately, they took the sample card back with them...)  The card
will digitally sign a challenge from the host, so you can't substitute
a bogus card.  The cards have non-volatile onboard storage for users'
secret keys and for the public keys of a certifying authority.

They are building a library/API for Tessera, called Catapult, that
will provide an interface suitable for many different applications.
They have prototype email and ftp applications that already uses it.
They intend to eventually give away source code for this library.
They responded favorably to the suggestion that they put it up for
anonymous ftp.

Applications (which can use the library and which the NSA approves for
government use) will be responsible for managing the LEAF field.  Note
that they intend to apply key escrowed Skipjack to other applications,
including mail and file encryption.  The LEAF would be included in
such places as the mail header or the file attributes.  This implies
that it is possible to omit sending the LEAF -- but the decrypt chip
won't work right if it doesn't get one.

When asked, they indicated that it might be possible wire up a pair of
Clipper/Capstone chips to not transmit the LEAF field, but that the
way to do this is "~not obvious from the interface we give you~" and
"~you'd have to be careful not to make mistakes~".  They gave a lot of
attention to obvious ways to get around the LEAF.

The unit key is generated via Skipjack itself, from random seeds
provided by the two escrow agencies (approximately monthly, though
that isn't certain yet).  They say they prefer a software generation
process because its correct behavior is auditable.

Capstone (but not Clipper) could be configured to allow independent
loading of the two key halves, in separate facilities.  "~It's your
money [meaning American taxpayers].~"

The LEAF field contains 80 bits for the traffic key, encrypted via the
unit key in "~a unique mode <grin>~", 32 bits for the unit id, and a 16
bit checksum of some sort.  (We didn't waste our breath asking what the
checksum algorithm was.)  This is all encrypted under the family key
using "~another mode <grin>~".

They expressed a great deal of willingness to make any sort of
reasonable changes that vendors needed for their products.  They are
trying *very* hard to get Skipjack and key escrow into lots of
products.