An unusual piece of cryptography history has gained an answer. Anthropic says Claude reconstructed how the original seed values behind several widely used NIST elliptic curves were generated, resolving a question that had attracted suspicion for more than two decades.
TL;DR: Claude identified a process involving the C library’s lrand48 generator and Unix timestamps. Anthropic stresses that the research does not reveal an attack on P-256 or other affected curves.
The seeds have mattered because the standards published their values without a convincing public account of how the numbers were chosen. The U.S. National Security Agency had supplied them, leaving generations of researchers to ask whether a secret selection method could hide some mathematical advantage.
Anthropic reports that a model worked backward from the published values and found an ordinary pseudorandom generation procedure. In the case of P-256, it took 1,842 iterations through candidate outputs to reproduce the standardized seed. The work also examined P-192, P-224, P-384 and P-521.
This is interesting to crypto users because elliptic-curve signatures underpin many kinds of account security, including passkeys and hardware-backed authentication. Explaining the provenance of parameters can improve confidence in the standards without changing the mathematics used by existing products.
There is a crucial limit to the story: recovering the seed-generation history is not the same as breaking a private key, proving the curve has a vulnerability or establishing that every implementation is safe. Anthropic itself says there is no urgent corrective action to take.
The headline is less about AI defeating encryption than AI helping researchers reconstruct a forgotten design decision. In cryptography, knowing why numbers look the way they do can be almost as valuable as knowing that the calculations work.
Written by the News Desk and edited by Samuel Rae.