Claude traced the mysterious NIST elliptic curve seeds to a 1990s random number generator

For 25 years, a small set of numbers at the heart of modern encryption had no public origin story. The elliptic curves NIST adopted around 2000 for ECDSA, the most widely used elliptic curve signature standard, were generated from seed values the NSA supplied without explaining how it picked them. The usual story was that an NSA mathematician produced them by hashing sentences he later forgot. Many cryptographers were not satisfied, suspicion of a hidden backdoor never fully went away, and three years ago a cryptographer put up a public bounty for anyone who could explain them.

Anthropic's Frontier Red Team now says an internal Claude research model answered the question in under half an hour. Several people at the company had already chased the forgotten-sentences theory without success. The model instead spotted a statistical quirk in the seeds, roughly a 1 in 5,810 chance of appearing by accident, which pointed to lrand48, a basic random number generator in the standard C library of the time. From there it worked backward to the inputs: Unix clock timestamps from between May 1997 and April 1998.

Anthropic is explicit that this is not a vulnerability. Nothing about the security of the curves changes, and a few pieces of the story are still unexplained. The same week, Anthropic also reported that a Claude model working in a loop with an automated evaluator designed quantum circuits that beat the best published designs on most of the tasks it tried.

Our take

The crypto detail matters less than the speed. AI is starting to settle expert arguments that sat open for decades, in an afternoon. Expect a run of solved puzzles in old standards, old code and old data.