Anthropic Replicates P-256 Seed Generation Path, Matches 1997 Random Output

Anthropic has replicated the public seed generation path for the P-256 curve that could be used in Ethereum (ETH) passkey wallets. The researchers obtained the same value as a random number generator initialized with a 1997 timestamp, though they did not uncover the criteria for selecting the curve candidate or the base point generation path.
In their October 16 research release, Anthropic stated that Claude tracked the seed generation path for the elliptic curve standard provided by the U.S. National Security Agency (NSA). The P-256 seed matched the 1842nd candidate from a C library random number generator 'lrand48' initialized at 10:52:34 AM KST on August 16, 1997. They also published the replication code.
This result does not imply that Ethereum wallet private keys have been exposed or that new vulnerabilities were discovered. The seed is an open design value used to define a single curve coefficient, distinct from the private key that approves transactions or user signature credentials. Anthropic emphasized they found no new attacks or weaknesses in the curve and that users do not need to take any action.
The link between passkey wallets and P-256 is defined by Ethereum Improvement Proposal EIP-7951, which specifies a 'P256VERIFY' operation for validating secp256r1 signatures. The proposal supports signatures created on devices like Apple Secure Enclave, Android KeyStore, and FIDO2/WebAuthn devices, enabling wallet implementations that sign using keys stored in devices and verify them on Ethereum.
However, P-256 differs from the curve Ethereum has traditionally used (secp256k1). EIP-7951 is a proposal for validating P-256 signatures only, not implying all Ethereum wallets or existing keys are affected in the same way. The security discussion around passkey-integrated wallets should be distinguished from overall Ethereum wallet safety.
While Anthropic's replication explains how the seed came from a specific random number generator and timestamp, they noted that 876th candidate in the same run also produced a valid curve. It remains unclear why the 1842nd value was chosen over earlier candidates. The base point generation method for the curve also remains unresolved; Anthropic explained some paths for curves generated in May 1997 but did not uncover the base point generation for P-256, along with P-224, P-384, and P-521.
Anthropic clarified that their replication of seed sources does not fully restore the entire curve parameter selection process.
Korean Source
This article is an English localization of a Korean-language crypto news report. Original headline: 앤트로픽, P-256 시드 생성 경로 재현…1997년 난수 출력과 일치