Claude cracks 25-year mystery behind NIST's elliptic curves
What's the deal? Anthropic's Claude has solved a long-standing cryptographic puzzle: the origin of the "seeds" used to generate the NIST elliptic curves behind ECDSA, one of the most widely used cryptographic standards. In October 2026, an Anthropic employee asked an internal research model to figure out how the seeds were made — without sharing any hints. Claude cracked it in under half an hour.
Some background: ECDSA was adopted by the National Institute of Standards and TechnologyDealroom has a profile for this one. Try Dealroom → in 2000 as part of the FIPS 186-2 Digital Signature Standard. The seeds for its curves were supplied by the National Security AgencyDealroom has a profile for this one. Try Dealroom → with no explanation, fuelling years of suspicion that they might hide deliberate backdoors.
The twist: The lore held that NSA mathematician Jerry Solinas generated the seeds in 1997 by hashing a humorous passphrase he later forgot. That turned out to be wrong. The seeds came from the drand48 random-number generator, seeded with a Unix timestamp. Solinas died in 2023, leaving the question unanswered.
How it worked: The algorithm passed the current Unix timestamp to the generator and drew 20 bytes at a time as candidate seeds, rejecting invalid curves. P-256 required 1,842 iterations to produce its published seed.
What this means: This is not a cryptographic break or a new weakness. "In fact, this makes us more confident the curves were chosen without any backdoors or intentional weaknesses," Anthropic said. The seeds were made with a common pseudorandom generator initialised from the clock — something anyone can now reproduce.
A team of 11 Anthropic researchers claimed a bounty on October 15, 2026, that cryptographer Filippo Valsorda had announced three years earlier. They donated the prize to the Internet Security Research GroupDealroom has a profile for this one. Try Dealroom →, which operates Let's EncryptDealroom has a profile for this one. Try Dealroom →, tripling its value to $36,864.
One loose end: The method yields roughly twice as many valid seeds as were published, and researchers have yet to explain why. One theory is a bug in 1997-era code; calculations that now take minutes were state-of-the-art at the time.
The signal: The find closes a quarter-century of speculation and shows AI being turned loose on open problems in security research — not to break systems, but to document how they were built.
Read more: anthropic.com, words.filippo.io
Image credit: MDGovpics
