Claude’s ECDSA Investigation Solves a Longstanding Mystery
The Claude ECDSA investigation traced the origins of seeds used by several NIST elliptic curves. Anthropic says the seeds for P-192, P-224, P-256, P-384, P-521, and related ANSI curves were not hashes of a humorous phrase. Instead, they came from the lrand48 pseudorandom number generator found in 1990s-era C libraries, initialized with Unix timestamps.
What the ECDSA Finding Means for Security
By recognizing patterns from the old linear congruential generator, the researchers recovered internal states and regenerated the complete seeds. This is not an attack on ECDSA, and it does not require users to replace keys or change deployments. Rather, a simple and reproducible generation process makes deliberate insertion of a weakness less plausible and increases confidence in the parameters.
Important Open Questions Around the Curves
Claude also found clues for how base points of six early curves were generated, matching a random-number procedure in FreeLIP and related integer-arithmetic code. However, the base points for P-224, P-256, P-384, and P-521 remain unexplained, as do those of the NIST binary curves; the B-163 seed also resisted recovery. A few viable candidate curves were skipped, but the selection criteria are unknown. The result is therefore a major provenance advance, not a complete resolution of every historical question.
The key limitation is that reproducing timestamps and PRNG output explains how the values could have been produced, but cannot prove that every historical choice was free of human judgment. The available evidence also shows no exploitable backdoor in the curves. Anthropic said the discovery bounty will be donated to the Internet Security Research Group.