An internal Anthropic model has traced the decades-old mystery behind the Elliptic Curve Digital Signature Algorithm curve seeds to a simple source: timestamps. Anthropic said Claude identified how the seeds for NIST P-192, P-224, P-256, P-384 and P-521 were generated in under half an hour, ending years of speculation over numbers that had sat in the standard without explanation.
The result matters because the curves were adopted by NIST in 2000 as part of the FIPS 186-2 Digital Signature Standard, and their seed values became part of the record. For P-256, the published seed is SEED = C49D3608 86E70493 6A6678E1 139D26B7 819F7E90. Anthropic said the seeds were not produced by any hash algorithm. Instead, they came from the C library’s lrand48 random-number generator, initialized with a Unix timestamp and sampled over repeated iterations until the published value appeared.
That is why the finding landed now. In October 2026, a curious Anthropic employee asked an internal research model whether it could figure out how the seeds were generated without being shown the clue. Claude did the rest, and Anthropic said the work took under half an hour. The company also said the discovery does not amount to a cryptographic break, a new attack or a weakness in the ECDSA curves.
The seed question had lingered because the specifications listed a seed parameter with no explanation or justification. That silence fed concern that the curve choices might hide deliberate backdoors or intentional weaknesses, and Neal Koblitz and Alfred Menezes later captured that debate in a summary titled A Riddle Wrapped in an Enigma. Anthropic said its finding makes those suspicions much less plausible, but it also leaves a basic awkwardness in place: the seeds had originally been provided by the NSA, without a public explanation for who chose them or why.
One detail helps show how mechanical the process was. Anthropic said the algorithm drew 20 bytes at a time as a candidate seed, repeating the process until it matched the standardized value; P-256 took 1842 iterations. The company said several Anthropic employees had tried and failed to solve the puzzle before Claude did. Jerry Solinas, a longtime NSA employee who had been tied to a rumor about a humorous passphrase, was also part of the story, but that rumor turned out to be wrong.
For users and researchers of ECDSA, the immediate consequence is narrow but important: a long-standing provenance question has a plausible answer. Anthropic said there is no action to be taken or anything to do differently. What remains unanswered is not whether the seed values can be reproduced, but why the original documentation left the process unexplained for so long.
