Forwarded message:
Could you forward this to cypherpunks for me? Walter is a technoid and happened to be one of the plaintiffs in Steve Jackson Games (e-mail user on Illuminati). I'm hoping one of the 'punks might be able to comment to him on his theory -- it sounds interesting to me, but I'm not technical enough to know if its got any merit. Thanks.
Date: Thu, 21 Oct 93 18:05 EDT From: Walter Milliken <milliken@BBN.COM> Subject: Clipper chip stuff To: ssteele@eff.org
Hi Shari -- it's been a while....
I know you're probably not the right person to talk to about Clipper chip stuff, but I figured you could forward this and tell me who is.
Basically, I think I see NSA's backdoor in the whole scheme. Probably other people have noticed this too, but in case they haven't (and I haven't seen any comment on these lines on comp.org.eff.talk), I figured I'd point it out.
The trick isn't necessarily in the algorithm (which I don't know, of course). I think it's in the key generation process. I read Dorothy Denning's description as posted to comp.org.eff.talk the other day, and decided it sounded fishy. Why are the secret per-chip keys generated from the chip serial number (which is observable by anyone with the law-enforcement key)? To escrow keys, all you need is to 1) associate a chip serial number with a secret key and 2) split the secret key into two, unusuable parts for the escrow agents. I can't see any reason why the secret key isn't just a random number (or rather the XOR of two random numbers, one for each escrow agent).
Instead, we've got this complex algorithm for converting the chip serial number into the chip secret key. Thus, if you know the chip serial number, the key generating algorithm, and the initialization states for the key generator (provided by the two escrow agents) *you can compute the chip secret key*! (Or so it appears on a very superficial reading.)
I suspect you could dispense with knowing the initialization states, if you were NSA and could obtain any chip from the same batch -- you open the chip up (I'm sure *they* know how to do that), extract its secret key, and reverse-engineer the initialization data. (This last step is non-trivial, but I'd be surprised if NSA couldn't do it -- it's a variant of known-plaintext attack.) It's also possible that NSA may *supply* the initialization states, or an algorithm for generating them, or at least advice on how to pick them. After all, it's their game, they get to make the rules....
In any case, I can see no way that using the *observable* chip serial number to help generate the chip secret key can in *any* way improve the security of the system. You'd be much better off just sticking any old random number into the chip as a secret key, and just noting it down with the associated chip serial number in the escrow files. Personally, I think I'd use a non-algorithmic mechanism for generating random keys -- perhaps a truly random number source, such as atomic decay processes. There's only one of these things -- it can afford to get fancy.
Disclaimer: I'm not a cryptography expert, so it's possible I'm missing something here. Possibly factoring in the chip serial number makes it harder to crack secret keys if you somehow manage to obtain a few chips from the same batch and open them. But it certainly seems suspicious to me....
---Walter
-- -=> mech@eff.org <=- Stanton McCandlish Electronic Frontier Foundation Online Activist & SysOp "A nation that is afraid to let its people judge the truth and falsehood of ideas in an open market is a nation that is afraid of its people." -JFK NitV-DC BBS 202-232-2715, Fido 1:109/? IndraNet 369:111/1, 14.4V32b 16.8ZyX