Claude reverse-engineered NIST's 25-year-old ECDSA curve seeds in 30 minutes.
ECDSA underpins HTTPS, $BTC, and SSH. It relies on elliptic curve math where your public key is a point derived from adding a base point to itself k times. Your private key is k. The curve parameters—equation, coefficients, base point—were standardized by NIST in the 90s using seeds supplied by the NSA.
The NSA never disclosed how those seeds were chosen. They claimed randomness, but the crypto community suspected a potential backdoor: the NSA could have searched for seeds that create exploitable curves only they understand.
Claude cracked it. The seeds weren't cryptographically random. They came from a basic RNG seeded by Unix time. Every NIST seed mapped to a specific second between May 1997 and April 1998.
This confirms decades of paranoia around NIST curves wasn't unfounded. If an LLM can reverse-engineer the seed generation process in 30 minutes, state actors have known this for years. Any system still using NIST P-curves without migration plans is carrying unquantified tail risk.
Claude even paused mid-analysis thinking it was a trick question. The weakness was that obvious.
ECDSA underpins HTTPS, $BTC, and SSH. It relies on elliptic curve math where your public key is a point derived from adding a base point to itself k times. Your private key is k. The curve parameters—equation, coefficients, base point—were standardized by NIST in the 90s using seeds supplied by the NSA.
The NSA never disclosed how those seeds were chosen. They claimed randomness, but the crypto community suspected a potential backdoor: the NSA could have searched for seeds that create exploitable curves only they understand.
Claude cracked it. The seeds weren't cryptographically random. They came from a basic RNG seeded by Unix time. Every NIST seed mapped to a specific second between May 1997 and April 1998.
This confirms decades of paranoia around NIST curves wasn't unfounded. If an LLM can reverse-engineer the seed generation process in 30 minutes, state actors have known this for years. Any system still using NIST P-curves without migration plans is carrying unquantified tail risk.
Claude even paused mid-analysis thinking it was a trick question. The weakness was that obvious.