Claude helped reconstruct how NIST elliptic-curve seeds were generated
TL;DR: Anthropic reports that an internal Claude research model traced several public elliptic-curve seeds to a 1990s C random-number generator initialized from Unix timestamps.
The interesting part is the verification path: a recognizable bit pattern, recovery of generator state, and short code that reproduces the published values. For P-256, the reported timestamp and candidate number regenerate the exact seed.
This concerns the provenance of public parameters; ECDSA security is unchanged. The report also identifies gaps: B-163's seed, several base points, and the reason some viable candidates were skipped remain unexplained.
A concrete example of AI-assisted investigation with outputs that people can check directly. Published October 16.
