I’m sure many of you have heard this urban legend about the NSA having forced NIST standardization of ECDSA with weakened seeds in order to give them an advantage at brute-forcing private keys or even possibly a backdoor.
The mystery was just solved by Anthropic Claude:
The discovery helps answer a question that has lingered for decades: whether the U.S. National Security Agency (NSA) deliberately chose those numbers to introduce hidden weaknesses.
[fact-checking!] NSA’s darkest cryptographic secret of the 1990s: it seeded its random numbers like a first-year student. Claude tested the decades-old theory of some cryptographers that the NSA secretly planted weak NIST elliptic curves behind unexplained seeds.
The results pointed to lrand48, a generator available in C programming libraries, initialized with Unix timestamps, numbers representing dates and times.
For the five NIST curves, the recovered timestamps fall between May 1997 and April 1998. For example, the P-256 seed was reproduced using August 16, 1997, at 01:52:34 UTC and the 1,842nd candidate generated.
So, for once at least, they were not the bad guys - they were just dumb
Perhaps it doesn’t explain why NSA couldn’t just publicly explain, at the time, how they came up with the magic numbers. Then there wouldn’t have been a mystery to explain / myth to debunk.
I’m not even sure that they are “dumb”. As the article says,
A predictable generator would be dangerous if used to create private keys. Here, however, it produced starting material for mathematical parameters that were always public.
So it’s basically just starting with a Unix timestamp and doing some key stretching in order to give a magic number for another algorithm.
Or it is an instance when Claude found predefined numbers in a stream of random numbers — in one of many candidate streams of random numbers — and claimed that it is definitely where these numbers came from. It could as well have found these numbers at a certain place among digits of π, because, as we know, we can find any finite digit sequence in there. This is not the first instance when an LLM produced a plausible nonsense.
I would believe it if there was a proof of improbability of finding a random digit sequence of required length among these candidate sequences near their beginning.
It is just their normal mode of operation to stamp “Top Secret” markings on anything they do regardless of whether it makes sense. They once classified their partial results on solving a publicly available cryptographic riddle made by some artist.
I know the news Vuntun, but it will be only be some kind of truth if some other independent LLM, News Paper, Human… reviewed the sources, math and subscribe this kind of “calculation”. If it does not, its just some true or hallucination of a child, dreaming human or want to be-leaver social media group. I do not say fake news but you know… like gossip.
The part of your sentence after “as we know” is in fact not known.
It could be true. It is conjectured as being true. But noone has proved it to be true.
Because π is irrational, its decimal expansion cannot be finite and therefore is infinite and, being infinite, therefore cannot be repeating.
So the simple question becomes: if I have an infinite non-repeating stream of digits, does it necessarily contain all possible finite digit sequences? The answer is “no”. Proof by counter-example. Several different ways. EFTR.
The much harder question is: for the specific decimal expansion of π does it contain all possible finite digit sequences? The answer is at this time not known (well unless someone has proved it recently).
Let’s suppose for the moment that the conjecture is true. In that case, using the digits of π for crypto algorithm magic numbers would be slightly dangerous i.e. would not dispel the “myth” that the NSA did something dodgy in the choice of the magic numbers.