For more than thirty years, cryptographers have puzzled over a quiet detail buried in the specification of NIST's standard elliptic curves, including P-256, the curve underpinning much of modern ECDSA signing. Each curve ships with a seemingly random 'seed' value whose origin was never documented, fueling long-running speculation that the US National Security Agency may have chosen these numbers deliberately, potentially to embed a hidden weakness. No one had ever been able to show what process, if any, produced those seeds.
According to Anthropic, that mystery has now largely been resolved with help from Claude. Working alongside independent cryptography researcher Filippo Valsorda, Anthropic's team used the model to search for plausible input strings that, when run through SHA-1, would reproduce the published seed values exactly. This is a classic needle-in-a-haystack problem: the search space of candidate phrases is enormous, and only a correct preimage will match the known hash output bit for bit.
The reported outcome is that Claude helped surface candidate inputs tied to mundane, traceable sources rather than anything suggesting a secretly engineered weakness. Anthropic frames this as evidence that a general-purpose language model can meaningfully assist with narrow, highly technical research tasks in cryptography, a field where exhaustive manual search has historically been the only option.
The practical security implications are limited, since P-256's resistance to known attacks does not depend on the seed's backstory, and no credible cryptanalysis has ever been built on the backdoor hypothesis. The real significance here is more about method than outcome: a decades-old open question in applied cryptography appears to have made progress not through a new mathematical insight, but through AI-assisted historical and combinatorial search, a pattern likely to recur in other poorly documented standards.
It is worth noting that the result comes from Anthropic's own research blog, describing work undertaken in collaboration with Valsorda, whose separate write-up is the original source of the investigation. Independent verification and peer scrutiny of the exact preimages found will be needed before the mystery can be considered fully closed.