Anthropic said on October 16 that an internal Claude research model traced five NIST elliptic-curve seeds to output from a 1990s random-number generator initialized with Unix timestamps. The result makes deliberate weakening less plausible, but it does not rule out every concern.

This is not a cryptographic break, and Anthropic says users do not need to change anything.

How Claude matched the seeds

The model traced the seeds for five curves to output from lrand48, a pseudorandom number generator used in older C libraries:

  • P-192
  • P-224
  • P-256
  • P-384
  • P-521

Anthropic said Claude found the matches in under half an hour. For P-256, the published seed appeared as the 1,842nd candidate in the reconstructed sequence. Anthropic also published code so readers can reproduce its results.

NIST’s current elliptic-curve guidance lists these curves and their parameters. A curve seed helps define a parameter used to specify the curve. Tracing how a seed was generated does not reveal users’ private keys or break their signatures.

What the finding leaves open

The seeds had drawn suspicion because their origins were not explained when the parameters were published. Anthropic says the simple, reproducible generation process makes deliberate weakening less plausible. That is a reason to worry less, not proof that every concern has been eliminated.

Several choices remain unexplained. Anthropic could not account for the B-163 binary-curve seed or how the base points for P-224, P-256, P-384 and P-521 were chosen. It also said some candidate seeds that could produce viable curves were skipped, without identifying the selection criteria.

Anthropic’s work offers an explanation for five curve seeds, but not for every choice behind the parameters. The company says the findings reveal no weakness and require no user action.