Anthropic says Claude traced NIST's ECDSA curve seeds to a pseudorandom number generator, challenging decades-old suspicions of an NSA backdoor in the algorithm

The origins of widely used cryptographic seeds are now known, settling long-standing suspicions of intentional backdoors.

Anthropic's AI model, Claude, has identified the origins of the seed parameters used in several widely adopted Elliptic Curve Digital Signature Algorithm (ECDSA) curves, including NIST P-192, P-224, P-256, P-384, and P-521. For decades, these seeds, provided by the NSA, were a subject of suspicion due to their unexplained provenance, fueling speculation about potential backdoors. Claude discovered that the seeds were not the result of cryptographic hashes as previously theorized, but rather outputs from the C library's `lrand48` pseudorandom number generator, initialized with Unix timestamps from the mid-to-late 1990s for prime curves and early 1980s for binary curves. Additionally, Claude identified the origins of base points for some of these curves, tracing them to a random number generator in the FreeLIP library. This resolution aims to close years of speculation within the cryptographic community.

Claude's discovery that NIST ECDSA curve seeds originated from a simple pseudorandom number generator initialized by Unix timestamps significantly reduces the plausibility of intentional backdoors. This revelation transforms the mysterious NSA-provided seeds into an artifact of common programming practices from the 1990s, making the parameters appear more innocuously generated and bolstering confidence in their security. The ability to reproduce these seeds using publicly known methods reassures users that the foundational elements of ECDSA are not compromised by hidden vulnerabilities.

The successful identification of the seed origins by Claude highlights the growing capabilities of AI in unraveling complex, long-standing technical mysteries. While the findings do not represent a cryptographic break, they provide a crucial piece of evidence for the integrity of widely used security standards. The next steps involve community verification of these findings and further analysis into the selection criteria for the base points of the remaining NIST curves, which remain unexplained.