Post-quantum RSA
eprint.iacr.org
eprint.iacr.org
a nice, thought-provoking paper that is nevertheless purely speculative, and also more than a little tongue-in-cheek in parts. (Note that one of the keywords is "Make RSA Great Again".) It's basically pointing out that there is a small gap between the costs of legitimately using RSA and factoring via Shor's algorithm. With truly massive keys, this gap can be exploited to provide concrete security guarantees for RSA even in the presence of fast quantum computers.
THIS PAPER IS NOT:
a revolutionary new quantum factoring algorithm. They show how to apply quantum Grover search to a particular step of a known factorization algorithm based on elliptic curves.
QUANTUM COMPUTING DOES:
break public-key cryptosystems based on the hardness of factoring and finding discrete logarithms. Additionally, it gives a quadratic speedup in generic search problems (via Grover search) which effectively halves the key size of symmetric encryption schemes and yields faster algorithms for things like finding collisions in hash functions.
QUANTUM COMPUTING (PROBABLY) DOES NOT:
break all asymmetric cryptography. There is no known proof of this, but it's widely believed that asymmetric cryptosystems based on different hardness assumptions are not vulnerable to (non-generic) attacks by quantum computers.
My favorite quote from the paper on that topic is
"Something expensive might nevertheless be useful if it is the least expensive way to achieve the user’s desired security goal"
Post quantum crypto isn't the end of crypto, is the end of the meme that nation-state-proof cryptography is "too cheap to meter" (A nuclear power phrase from the old days)
The pre-quantum landscape is cheap unbreakable algos implemented in easily breakable ways as a part of easily breakable systems, so there's quite a bit of criminal activity going on out there. The post-quantum landscape is very expensive unbreakable algos implemented by professionals because the cost (computational and financial) is large. The involvement of professionals indicates that criminal activity based on breaking crypto implementations is likely to drop.
Its likely that Hollywood style multiple levels of crypto will soon exist. The days of being able to keep the NSA out of your wifi access point will be over unless you're willing to spend ridiculous amounts of money. However, for a very modest expense it'll still be possible to keep your neighbor off your wifi, wide open to nation states, sure. But my wifi leaching neighbor is a more realistic adversary than the entire NSA.
Hell, there's already work out there (http://eprint.iacr.org/2014/599.pdf) for a quantum-resistant key exchange algorithm that's fast enough to have only 21% less throughput than traditional elliptic Diffie-Hellman.
The trick in the big-key RSA paper is knowing how Shor works, because the average big-O is low for a randomly selected large number, but the maximal number of ops that could be required is extremely large and yet that number is "easily" constructible. So merely intentionally and methodically (and expensively) manufacture a "poison pill" composite that Shor can't possibly factor because that number was constructed specifically to make Shor algo fail, and you're RSA system will be good.
The algo in your linked paper is very convenient. However its newer and not as seasoned and the wolves are snapping at its heels there was that paper around Thanksgiving last year (which did get retracted) claiming to partially quantum break LWE.
https://arxiv.org/abs/1611.06999
I'd like to see a proof that there exists no possible theoretical quantum speedup against LWE. Yes I know there isn't one in currently described or implemented, I mean I can't find a proof that there cannot exist an attack against LWE. Or, a proof that a non-quantum attack against LWE cannot exist because in the end that would imply 1=2 or something invalid so by induction its unbreakable.
In the year 3000 I think the odds are good that algos will still have a large ratio between the average ops when run against a random number vs the maximal ops when run against a constructable poison pill designed to require maximum number of ops. Yet in the year 3000 I would not put faith in any one algo that hadn't been broken by mid-2017 will never be broken. It seems better to bet on math as a field continuing to have algos with a wide spread, than to bet on any one algo.
In that way I think it likely we'll get stuck with something like big key RSA over a long enough time period.
While at the same time I can agree with you, yes in the post-quantum world starting in '94 and LWE starting in 2005 there was a post-quantum era from 2005 to at least mid 2017 where LWE was a fast unbroken post-quantum system. Will that end-date continue to 3000 AD?
Right now I'm thinking about if the Shor trick could be applied to a theoretically quantum broken RLWE-KEX. Random Discrete Gaussian Sampling is random... Is it possible to calculate poison pill samples that would mess up a theoretically broken RLW-KEX (depending a lot on the characteristics of an imaginary break?) I found slides by Peikert online about how those samples are selected, which I haven't made sense of yet.
I use the term quantum-resistant, which I think is more clear: crypto for which Grover's algorithm or Shor's algorithm cannot efficiently break it.
> The algo in your linked paper is very convenient. However its newer and not as seasoned
Totally agree, but it does show that you shouldn't just assume quantum-resistant crypto is necessarily computationally expensive.
> I'd like to see a proof that there exists no possible theoretical quantum speedup against LWE.
Don't hold your breath :). We don't even have a proof that integer factoring can't be broken in polynomial time in classical computers. We don't even know if one can make public key crypto unbreakable to classical computers (equivalent to P vs. NP problem). And the relationship of BQP (problems solvable in polynomial time on a quantum computer) to P and NP is still an open problem. It could turn out BQP=P, and quantum computers don't even give an asymptotic speedup over classical computers (I don't think that's likely, but it hasn't been proven one way or the other).
The more important point here, though, is that there's nothing particularly "expensive" about this stuff. It's all slower than RSA and curves, but (in particular) curves have gotten much faster over the last 10 years, and obviously computers are getting faster and silicon offloads less expensive, so the performance setback here isn't that big.
I'm also not sure why you have so much confidence in the conventional strength of RSA.
It's certainly not the case that PQ crypto requires enormous investments. Google ran a long quiet experiment where they used RLWE alongside ECDHE for users, and none of those users even noticed.
Am I blind or did they not include the actual algorithm in the paper? I'm not very knowledgeable in this sort of stuff so to be fair I was skimming for just a simple writeout of the algorithm, I may have just missed it.
Isn't that also huge news, depending on the definition of "often"?
I wonder if a quantum computing revolution would cause a resurgence of research into symmetric encryption algorithms?
Because doesn't Shor's algorithm basically destroy asymmetric encryption (i.e., integer-factorization schemes like RSA are solvable in polynomial time), while symmetric encryption is only weakened? Namely, I think I remember reading that a quantum computer can search through a space of size 2^n in time 2^(n/2) for AES & friends.
I believe you're thinking of Grover's algorithm, which as you remember offers the ability to search for a value in a function domain in O(sqrt(N)). This is trivially countered just by doubling the key length, a 128-bit key probably couldn't be considered secure (average 2^64 evaluations) but a 256-bit would still average 2^128 and 512-bit 2^256. I vaguely recall there was some speculation that applied de Broglie–Bohm theory, if it was accurate, could allow a mild improvement but even then it only changed it to cube-root, which again could be countered with slightly larger keys (in practice 512-bit would still be plenty even then).
At least for now there don't appear to be any major algorithmic implications for symmetrical encryption from a theoretical future scalable general purpose quantum computer. Granted, there may be some practical historical ones, and in a few cases they could be of significance. If agencies have archived data of interest encrypted with 128-bit in the past, they may gain the ability to gain access to it within the next decade or two. A lot of it might be well past its proverbial sell-by date but not all if those using it were counting on it last much longer then they'd be alive.
EDIT: Also note that quantum cryptanalytic attacks may be able to break specific symmetric ciphers by exploiting some structure specific to the cipher. But you're correct that QCs don't generically break symmetric cryptography.
Then it constructs a 1 terabyte RSA parameter, so big neither algorithm can currently attack it using a computer (quantum or otherwise) of practical size, as all known quantum attacks (including the new presented) would still need over a 2¹⁰⁰ workfactor; the 2³¹ factors they would need to find are each 4096 bits long themselves, if I'm reading it right?
Needless to say, that is a trifle on the heavy side and definitely isn't your first practical choice (those would be lattice-, code- or hash-based primitives), but it is possible, and that's impressive (to me) by itself! Nice work.
Also, while new, this result isn't really surprising. That Grover's algorithm can be applied to get a quadratic improvement over existing classical search algorithms has been known for a long time.