The whole daySat 17 Oct 2026
AI Compass

Anthropic’s Claude finds five NIST curve seeds came from timestamps, not hashes

In a 16 October 2026 post, Anthropic says an internal research model showed in under half an hour that the unexplained seeds of five NIST elliptic curves came from a C library random number generator seeded by the clock.

A paper collage of a torn purple strip on a dotted pink background beside a striped sunflower seed on dark grey and green paper
Image: AnthropicAnthropic’s share image for the post

ECDSA, the most widely used elliptic-curve signature standard, was adopted by NIST in 2000 in FIPS 186-2. Its curves are partly defined by seed values that the NSA supplied without explanation, which fed years of suspicion that they could hide backdoors, and a bounty for whoever found the seeds’ origin.

An Anthropic employee asked an internal research model how the seeds were made, without pointing it to the rumoured clue that they were hashes of a humorous phrase. In under half an hour, Claude showed that the P-192, P-224, P-256, P-384 and P-521 seeds came from lrand48, the C library’s random number generator, seeded with Unix timestamps, readings of the computer clock, from May 1997 to April 1998.

Each of the five NIST prime-curve seeds is matched to a Unix timestamp and a candidate number: P-256’s seed is candidate 1,842 of a run started on 16 August 1997, and P-224’s is candidate 7,026 of one on 3 April 1998
Anthropic’s table: each seed’s timestamp t, that time in UTC and its candidate number nImage: Anthropic

Claude spotted a pattern in the seeds’ top bits, recovered the generator’s hidden state and checked that each recovered state regenerates all 160 bits of its seed, which does not happen by chance. It also explained the base points, the fixed point every public key on a curve is derived from, of P-192 and five ANSI X9.62 example curves.

Anthropic says this is not a break of, attack on or weakness in ECDSA, and that a simple, reproducible method makes deliberately weakened curves much less plausible. The $36,864 bounty will go to the Internet Security Research Group.

The seed of the binary curve B-163 and the base points of P-224, P-256, P-384 and P-521 remain unexplained, and Anthropic doesn’t know why some runs passed over workable curves.

P-256’s run passed over one good curve, candidate 876, before its published seed at candidate 1,842; P-224’s run passed over three, the two ANSI X9.62 runs seven more, and the P-384 and P-521 runs none
Anthropic’s chart of the workable curves each run passed over before its published onesImage: Anthropic

Sources

  1. PrimaryClaude reveals ECDSA curve originsanthropic.com · Anthropic Frontier Red Team post, 16 October 2026, with code to reproduce

The rest of the day

Read the whole day

Spotted an error? Email aiforsuccess.in+compass@gmail.com or tag @ai_for_success on X