Every time you unlock a phone with your face, log into a bank or send a private message, your device performs a small piece of mathematical magic in the background. Much of that magic rests on elliptic curve cryptography, a family of techniques built around carefully chosen numbers — curves, constants and starting values known as seeds. To most of us they are invisible. To cryptographers, they raise an awkward question that has lingered for decades: who chose these numbers in the first place, and can we be sure they were chosen honestly?

That question has always centred on the NIST curve seeds, the starting values behind several widely used elliptic curves standardised by the US National Institute of Standards and Technology. The seeds were supplied decades ago by the National Security Agency, and no full public explanation of how they were generated was ever provided. Nothing about them was ever shown to be wrong, but the absence of an explanation left room for doubt — because in cryptography, a constant nobody can account for is a constant nobody can fully trust.

Why Provenance Matters

A seed matters because it shapes everything built on top of it. If a curve’s starting values had been secretly chosen from a rare, weak subset of possibilities, whoever knew the trick might hold an advantage over everyone else’s encryption — the mathematical equivalent of a landlord quietly keeping a master key. This is why standards bodies now prefer “nothing up my sleeve” numbers: constants derived in a public, boring, reproducible way, such as the digits of pi, so that no hidden selection is even possible. The NIST seeds, generated in an earlier and less transparent era, never quite met that standard of openness.

Provenance, then, is not pedantry. It is the audit trail of the digital world’s foundations. When billions of devices rely on the same handful of curves, knowing where their parameters came from is part of knowing whether the whole structure deserves the trust placed in it. Researchers have picked at the question for years, testing theories about how the seeds might have been produced, without reaching a conclusive answer.

An AI Reopens a Cold Case

The story took an unexpected turn when Anthropic pointed its AI model, Claude, at the puzzle. Working backwards from the published values, Claude traced the NSA-supplied seeds to a decidedly unglamorous source: lrand48, an old pseudorandom number generator once common in Unix systems, combined with Unix timestamps. In the case of the widely used P-256 curve, the model reproduced the seed in 1,842 iterations — a short search that fits an ordinary, era-typical generation process rather than anything exotic. Anthropic is careful to stress what this does and does not mean: the work found no attack on the curves, and no weakness in the encryption people use every day.

Bitcoin enthusiasts may note that their network sidesteps the question altogether, since Bitcoin uses a different curve, secp256k1, rather than the NIST family. For the rest of the internet, though — banking, messaging, secure web browsing — the NIST curves remain load-bearing, which is why a small piece of mathematical archaeology attracted such attention.

The Reassuring Power of a Boring Answer

In security, boring is beautiful. A seed that turns out to have come from a standard library function and a timestamp looks like the product of a programmer using the normal tools of the time, not like a value engineered to hide a trapdoor. That does not prove every implementation of the curves is flawless, and researchers will keep testing them, as they should. But it replaces decades of speculation with a mundane, checkable origin story — and in a field built on verifiable trust, an explanation anyone can reproduce is worth a great deal. The numbers at the heart of internet security are a little less mysterious than they were, and the internet is slightly better off for it.