NIST adopted ECDSA in 2000 as part of the FIPS 186-2 digital signature standard. Alongside each curve over a prime field, the specification published a "seed" parameter whose hash value defines the curve coefficient. Those seeds arrived from the NSA with no explanation, and their provenance has been an open question in cryptography for decades. Plenty of researchers quietly suspected the numbers hid something. Anthropic says a curious employee simply asked one of its internal research models whether it could work out how the seeds were generated.
The answer arrived in under half an hour. The seeds behind P-192, P-224, P-256, P-384 and P-521 were not hashes of any phrase. They were output from the C standard library lrand48 pseudo-random number generator, initialised with ordinary Unix timestamps from 1997 and 1998. The base points of six earlier curves trace back to the random number generator inside FreeLIP, Arjen Lenstra's long-integer package. Anthropic published code so anyone can reproduce the finding, and some seeds and base points remain unexplained.
Anthropic stresses this is not a cryptographic break, attack or weakness. If anything it strengthens confidence that the curves were chosen without deliberate flaws, because they can now be regenerated from a clock reading rather than trusted on faith. Researchers are told to read it as a closing chapter, with no action required. For founders shipping products that lean on TLS and signing, the practical takeaway is simple: the standards underneath your infrastructure just got a cleaner lineage, and a research model earned its keep on a question humans had parked for thirty years.