Anthropic said on October 16, 2026 that Claude reconstructed how seed values were generated for five NIST prime curves used with ECDSA, addressing a provenance question that had lingered for about 25 years.
In the Anthropic research, the company says Claude matched the published seeds for P-192, P-224, P-256, P-384, and P-521 to a repeatable process using the C library PRNG lrand48 with Unix timestamps from 1997 and 1998. Anthropic also published code to reproduce the result.
Anthropic says this is not a cryptographic break, a new attack, or a weakness in ECDSA, and says users do not need to change anything. The claim is narrower: it offers evidence about how the curve parameters were originally produced.
Why it matters: suspicion around these curves centered on whether opaque seed generation might have hidden a backdoor. Anthropic argues the recovered process makes that less plausible for the curves it analyzed, even if some selection details remain unexplained.