Claude reveals ECDSA curve origins

Open original

Anthropic's red team reports that an internal Claude model traced the origins of the NIST elliptic curve seeds to the lrand48 random number generator seeded with a Unix timestamp, ending a decades-long question about whether the curves were backdoored. The team says the finding shows the curves were generated in a simple, reproducible way.

  • The NIST P-192, P-224, P-256, P-384, and P-521 seeds are outputs of the C library's lrand48 generator seeded by Unix timestamps from 1997 to 1998, not a hash of a humorous passphrase as rumored.
  • Claude recovered the generator state by spotting a pattern of zero top bits in the seeds, then solving a truncated linear congruential generator using lattice reduction (LLL).
  • Some base points, including those of the ANSI X9.62 curves, trace to the FreeLIP and NTL random number generators, but the base points for P-224, P-256, P-384, P-521, and the NIST binary curves remain unexplained, as does the B-163 seed.

NIST's elliptic curve standards from 2000 (FIPS 186-2) came with "seed" values that the NSA supplied with no explanation, and for about 25 years nobody could say where they came from. The seeds fed a hash that set the curve coefficient b, which made them the obvious place to look for a hidden backdoor. A bounty went unclaimed, a popular story said the seeds were a hash of a joke passphrase attributed to NSA employee Jerry Solinas, and Anthropic employees chased that lead without success. Solinas died in 2023 and his colleagues didn't answer questions, so the mystery stayed open. Anthropic says an internal research model, Claude, solved it in under half an hour in October 2026, after being asked to work it out without the passphrase hint. The seeds weren't hashes at all. They came from the C library's lrand48 generator, a weak 1990s linear congruential generator, seeded with a Unix timestamp. Claude noticed that the top bit of the output words was zero in a pattern that would occur by chance with probability about 1 in 5,810, which pointed to lrand48. For the prime-curve seeds, which kept only part of each output, it recovered the hidden generator state by treating it as a lattice problem and running the LLL basis reduction algorithm. The recovered initialization values map to timestamps between May 1997 and April 1998 for the prime curves, and to February 1984 for the binary curves. The binary-curve timestamps look like they were doubled, which would put them in April 1998, so the generating program may have halved the clock reading at some point. Claude also traced the base points of six of the curves to the random number generator in Arjen Lenstra's FreeLIP library. Anthropic's bottom line is that this is not a break of ECDSA. The curves were produced by a simple, reproducible process that anyone can now rerun, which makes a deliberately planted backdoor much less plausible than it looked before. Rerunning the process still shows a few things worth noting. Choosing parameters from a clock-seeded generator is a weak procedure by modern standards, even if it wasn't sinister. And the explanation rests on the seeds and the published generator behavior, not on a confession from the people who ran the code, so the account is strong on the mechanism but thinner on intent.

Ask about this article