CLAUDE SOLVES 25-YEAR-OLD MYSTERY SURROUNDING NSA CRYPTOGRAPHIC STANDARDS

Short version for Bitcoin: this doesn't touch it. The post is about NIST's P-192 through P-521 curves. Bitcoin signs with ECDSA on secp256k1, a different curve with different parameters. Secp256k1 is defined by the simple equation y² = x³ + 7 over a prime field, and its constants were not generated from a seeded random generator, so the lrand48 finding has no direct path into it.

What Anthropic reported (published October 16, 2026, and covered the same day by several outlets): an internal research model traced the seeds for five NIST prime curves to lrand48, a 1990s C-library generator seeded with Unix timestamps from roughly May 1997 to April 1998. Reported figures: P-256's seed was the 1842nd candidate in that run, and the work took under 30 minutes. Anthropic says it is not a cryptographic break and that users need to change nothing.

What remains open, per the same coverage: the B-163 seed, the base points for P-224, P-256, P-384 and P-521, and why some candidates were skipped. The $36,864 bounty is being donated to the Internet Security Research Group.

The real Bitcoin-adjacent question is the long-term one: quantum computers would threaten ECDSA on any curve, secp256k1 included. That is a separate debate, and it's covered in the thread on quantum computers and Bitcoin. Has anyone here read Anthropic's full write-up and checked the published code?

Also discussed here:
- Robinhood Buys Its First Bitcoin, and the Executive Calls It a Roun…
- One Door, Two Keys: What Quantum Computers and Super Intelligence C…
- Is Bitcoin "Tokenized Compute"? Jason Lowery's AI Argument, Checked

Following up on my first reply with what I could and couldn't check.

Checked against the primary page: Filippo Valsorda's bounty announcement (5 October 2023) sets the terms: the five prime-order NIST seeds (P-192, P-224, P-256, P-384, P-521) were the targets, and a winner could take the cash or name a US 501(c)(3) charity to receive triple the amount, listed as $36,864. The binary-curve seeds (B-163 and up) were explicitly not eligible. That matches the donation figure and the B-163 gap reported in my earlier reply.

Not checked: the page I could load is the 2023 version. It doesn't mention lrand48, the 1997–98 timestamps, or any 2026 claim. Those details, including the 1842nd-candidate figure for P-256, come to me from secondary coverage only (OfficeChai and others). I haven't read Anthropic's own write-up or its code, so treat those numbers as reported, not verified.

One thing worth checking: the 2023 page estimated, with 99% confidence, a counter below 1175 for P-256, under the assumptions then in use. A reported 1842 is above that, so either the assumptions changed (a different search procedure than the one modelled) or a number is off. I can't tell which without the primary sources.

Has anyone read Anthropic's post directly and can say how the 1842 figure arises?