Claude reportedly traces the origins of five NIST elliptic-curve seeds

Anthropic’s Claude reportedly traced the origins of five NIST elliptic-curve seeds to a simple pseudorandom-number generator, offering a concrete answer to a cryptography puzzle that has fuelled speculation for decades. The finding is not an attack on the curves, but it may make one particular backdoor theory less likely.

Anthropic Watch analysis

What happened

U.Today reports that Anthropic described an internal version of Claude finding how the seeds for P-192, P-224, P-256, P-384 and P-521 were generated. The account says the model traced them to lrand48, a pseudorandom-number generator found in older C libraries, seeded with a Unix timestamp. The seeds’ provenance had been debated partly because the NSA supplied them with little explanation.

The report says Claude found the origins in about half an hour. That is a claim about the reported research, not a general measure of how quickly AI can solve cryptographic problems. The article also notes that questions remain about the selection of base points and why other mathematically viable curves were skipped. Read U.Today’s report.

Why it matters

The result, as described, is not a successful attack on ECDSA and does not show that cryptocurrencies using elliptic-curve cryptography are unsafe. Instead, a reproducible account of how these particular seeds were generated could make deliberate weakening of these curves seem less likely. That distinction is doing useful work here; “AI solves cryptography” would be a rather less helpful summary.

It is also a striking example of AI helping investigate a technical mystery, rather than simply generating code or text. But solving one question about the seeds does not resolve every uncertainty about how the curves were selected.

Our read

This is an intriguing research claim with a narrower conclusion than the headline phrase “solving a cryptography mystery” might suggest. The reported finding could ease one long-running concern about these curves, but it is not a blanket security guarantee. Readers should separate the specific seed-generation result from broader claims about cryptography or AI’s ability to break it.

What to watch

• Whether Anthropic publishes further technical detail or reproducible evidence about the seed-generation method.
• What researchers conclude about the remaining questions around base-point selection and omitted curves.
• Whether independent technical scrutiny supports the reported account.

Discussion spark: Does a reproducible explanation for these five seeds meaningfully reduce your concern about the curves, or do the unanswered questions about their selection matter just as much?

Anthropic Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by Anthropic.