Anthropic Says Claude Found Timestamp-Based Origins for Five NIST Curve Seeds
Anthropic Says Claude Found Timestamp-Based Origins for Five NIST Curve Seeds
AI Research & Science

Anthropic Says Claude Found Timestamp-Based Origins for Five NIST Curve Seeds

AnthropicOct 16, 2026

Anthropic says Claude traced five NIST elliptic-curve seeds to the C library’s lrand48 generator, initialized with Unix timestamps—not a hash.

Anthropic reported on October 16 that an internal research version of Claude had identified how the seeds for five NIST elliptic curves were generated. The company says the model found that the values came from the C library’s lrand48 pseudorandom-number generator, initialized with Unix timestamps, rather than from a hash algorithm.

The seeds belong to the NIST P-192, P-224, P-256, P-384 and P-521 curves used in elliptic-curve cryptography. Anthropic says Claude reached the explanation in less than half an hour after an employee asked it to investigate the seeds’ origins without directing it toward the long-circulated theory that they came from a humorous passphrase. For P-256, the published seed was the 1,842nd 20-byte candidate generated in the run Anthropic reconstructed.

To identify the method, Claude searched for patterns consistent with older random-number generators, recovered their internal states, and checked whether those states reproduced the published values. Anthropic also reports matching the same generator pattern in five additional seeds from the ANSI X9.62 standard. The company published code intended to let others reproduce the seed calculations.

The finding addresses a longstanding provenance question about parameters supplied to NIST without an explanation of how they were chosen. Anthropic stresses that it is not a cryptographic break or evidence of a weakness in ECDSA, and says the results make deliberate backdooring less plausible. The report does not settle every question: Anthropic says it could not explain the B-163 seed or the origins of several curve base points.

Anthropic said the $36,864 bounty associated with finding the seeds will be donated to the Internet Security Research Group. The result offers a concrete example of an AI model helping investigate a difficult mathematical and cryptographic question, while the company’s own caveats leave the remaining unexplained parameters open for further scrutiny.

Industry Impact

AI-assisted cryptographic investigation

Cryptography, cybersecurity, and AI research

Overview

Anthropic’s report shows an internal version of Claude helping investigate the provenance of cryptographic parameters, a question relevant to researchers and security professionals assessing trust in standards. The result does not indicate a weakness in ECDSA or establish that the curves were backdoored; Anthropic says it makes deliberate backdooring less plausible. In the near term, the published code enables independent checks. Longer term, the case illustrates a potential research aid for complex forensic work, while unresolved parameters mean scrutiny is not finished.

Key Impact Points

Claude traced five NIST curve seeds to lrand48 initialized with Unix timestamps, offering a testable explanation for their origins.

Anthropic published code intended to let others reproduce the calculations, supporting independent verification.

The finding is not a cryptographic break; the B-163 seed and several curve base points remain unexplained.

Opportunities

Research validation, not a proven commercial opportunity

Cryptography and cybersecurity research

The clearest near-term opportunity is research validation: Anthropic’s published code gives cryptographers and security teams a basis for checking the reported seed calculations. The article does not establish demand for a commercial tool, a new product, or a revenue model, so a business opportunity should not be assumed. Organizations already reviewing cryptographic standards can use the finding to inform further provenance checks, while researchers can examine the remaining unexplained B-163 seed and curve base points. Any broader workflow or service would need to demonstrate reliable results and practical demand beyond this case.

Reference link

https://anthropic.com/research/ecdsa-curve-origins

Image Source Link

https://media.rnztools.nz/rnz/image/upload/s--w_m5ROF8--/t_kt-crop-16x10/w_800/f_auto/q_auto:eco/4JN09WS_AFP__20260602__boivin_notitle260602_npoLZ__v1__HighRes__PhotoIllustrationAnthropicFi_jpg

Community Discussion (0)

Loading discussions...