Anthropic says Claude cracked NIST elliptic curve seed mystery
Claude identified how the seeds for five NIST curves were generated in under 30 minutes; Anthropic's team won a bounty and is donating $36,864 to Let's Encrypt's operator.
Anthropic announced on October 16, 2026, that its in-house AI research model Claude determined how the seed values for the elliptic curves in a NIST standard were generated, resolving a question that had gone unanswered for more than 25 years. According to japan.cnet.com, Claude identified the generation method for the seeds used for five NIST curves in less than 30 minutes.
According to Anthropic, the seeds were produced with the lrand48 pseudo-random number generator in a C library, starting from an initial state derived from dates and times in 1997–98. The company published code that reproduces the original seed values. Claude also worked out how seeds were generated for five curves in another standard and for four other curves NIST adopted.
No change for users
Anthropic said the finding does not mean the cryptography has been broken, and that no new attacks or curve weaknesses were found. The company added that users do not need to change any settings.
Cryptographer Filippo Valsorda announced on October 16 that Anthropic's research team had won his bounty. The team chose the donation option, and $36,864 will go to the Internet Security Research Group, the nonprofit that operates Let's Encrypt.
NIST published the standard in 2000 with seed values used to define the curves for ECDSA; the seeds were provided by the NSA without an explanation of how they were generated. In 2015, former NSA staff member Jerry Solinas said he recalled the seeds being made by hashing a joking English sentence, but he could not recall the exact procedure and no old records were found. In 2023, Valsorda and others set up a bounty of $12,288, tripled if donated to a qualifying US charity.
Quick answers
What did Claude find out about the NIST curve seeds?
Anthropic said Claude determined the seeds were generated with the lrand48 pseudo-random number generator in a C library, using an initial state derived from dates and times in 1997–98, and published code that reproduces the original values.
Does this mean the cryptography is broken?
No. Anthropic said the finding does not mean the cryptography has been broken, that no new attacks or curve weaknesses were found, and that users do not need to change any settings.
Where did the bounty money go?
Anthropic's research team chose the donation option, and $36,864 will be donated to the Internet Security Research Group, which operates Let's Encrypt.