Claude has reportedly solved a cryptography mystery that remained unresolved for more than 25 years, tracing the origin of several elliptic curve seeds to a conventional random number generator initialized with timestamps. The discovery, published by Anthropic on October 16, could help address longstanding concerns about whether the US National Security Agency deliberately introduced hidden weaknesses into widely used cryptographic standards.
The finding matters to the cryptocurrency industry because elliptic curve cryptography underpins digital signatures, which allow users to prove ownership of crypto accounts without revealing their private keys. However, the discovery concerns the NIST P-256 family of curves, not a demonstrated vulnerability in Bitcoin or Ethereum. Those networks use secp256k1, a different curve.
Claude Traces the Seeds to a 1990s Computer Program
The mystery began with elliptic curve parameters supplied by the NSA and published without a clear explanation of how their original seed values were generated. Researchers questioned whether the seeds could have been deliberately engineered to create a weakness known to the agency but difficult for others to detect. Those concerns gained attention as other cryptographic standards came under scrutiny over their design and security.
According to Anthropic’s research, an internal Claude model identified the origins of seeds for five NIST curves: P-192, P-224, P-256, P-384 and P-521. The model reportedly completed the investigation in less than 30 minutes, approaching the problem by examining how software from the 1990s might have generated the values rather than relying on earlier theories about a secret phrase.
Claude identified lrand48, a pseudorandom number generator available in older C software libraries. The research indicates that the generator was initialized using Unix timestamps from 1997 and 1998, with candidate seed values produced from its output. For P-256, the published seed matched the 1,842nd candidate in the reconstructed sequence.
The result challenges an earlier account suggesting that the values had been generated by hashing a memorable phrase. Instead, Anthropic says the numerical patterns match output from a relatively ordinary, reproducible computer process. This gives researchers a concrete explanation they can test rather than relying on recollections about how the original parameters were created.
The discovery does not prove that every part of the original curve-generation process was free from deliberate manipulation. Anthropic also identified unanswered questions, including why certain viable candidate curves were skipped and how some curve base points were selected. The company says the evidence makes a deliberate backdoor in the examined seeds substantially less plausible, rather than claiming that every historical question has been settled.
What the Discovery Means for Bitcoin and Ethereum
The distinction between P-256 and secp256k1 is important for crypto users. Bitcoin and Ethereum use secp256k1 for their traditional account-signature systems, while P-256 is widely supported by modern devices and security standards. Apple’s Secure Enclave and WebAuthn passkey systems support P-256, creating compatibility challenges for crypto accounts that depend on secp256k1 signatures.
That difference has practical consequences for wallets and applications. Native hardware-backed signing and passkey integration can make authentication more convenient, but developers must account for the cryptographic algorithms supported by each device and platform. The newly reported seed discovery does not automatically remove those compatibility barriers or enable Bitcoin and Ethereum accounts to use P-256 signatures.
It also does not mean that existing Bitcoin or Ethereum private keys have been exposed. Anthropic describes the finding as neither a cryptographic break nor a newly discovered attack, and says users do not need to change their security practices because of the result. The separate question of how future quantum computers might threaten current public-key cryptography remains unresolved by this research.
For the wider crypto community, the more significant development may be the role AI can play in technical investigations. Claude’s reported result shows how a model can combine pattern recognition, software analysis and mathematical techniques to investigate a problem that resisted earlier attempts. Independent reproduction of the calculations remains important, particularly when conclusions concern security standards used across financial systems.
The takeaway is not that AI has proved all cryptographic systems safe. It is that an obscure, decades-old question about the origins of important curve parameters now has a specific, testable explanation. If independently reproducible, the result could reduce one longstanding source of suspicion around P-256 while leaving other questions about cryptographic design, implementation and future threats open.
About The Author
