AI Cracks 25-Year NSA Crypto Seed Mystery, Easing Backdoor Suspicions
An Anthropic AI model identified the mundane origin of cryptographic numbers provided by the NSA in the 1990s, which were long suspected of containing a secret weakness, confirming they came from a basic random-number generator.
Key takeaways
- An Anthropic AI model, Claude, solved a 25-year mystery by tracing specific cryptographic numbers supplied by the NSA to a common random-number generator.
- The discovery largely dispels long-held suspicions that the National Security Agency had embedded a secret 'backdoor' into widely used internet security standards.
- Bitcoin's underlying cryptographic curve, secp256k1, was never part of this particular controversy and remains distinct from the NIST curves in question.
An artificial intelligence model developed by Anthropic has finally resolved a quarter-century-old cryptographic enigma, identifying the source of specific numbers provided by the U.S. National Security Agency (NSA) in the late 1990s. These numbers, known as seeds, were used to generate elliptic curves, which are foundational to much of the internet's security infrastructure and protect digital assets like Bitcoin wallets.
For 25 years, cryptographers debated the origins of these seeds, fueling suspicions that the NSA might have embedded a hidden weakness, or 'backdoor,' into the standards. This concern intensified following the 2013 Snowden disclosures, which revealed NSA efforts to compromise cryptographic tools. However, the AI model, Claude, took less than half an hour to trace the numbers to an ordinary random-number program called lrand48, fed with timestamps from 1997 and 1998.
The Mystery Unraveled
The U.S. National Institute of Standards and Technology (NIST) adopted a set of elliptic curves in 2000, which had been supplied by the NSA. The agency provided these designs without fully explaining the derivation of some critical numbers used in their generation. These public seeds, which define the structure of the cryptographic curves, became a focal point of distrust.
Cryptographers, including Neal Koblitz and Alfred Menezes, explored the controversy in their 2015 paper, "A Riddle Wrapped in an Enigma." Security researcher Steve Weis also spent years pursuing an explanation, filing public-records requests to no avail. The NSA had ceased responding to inquiries, and NIST stated it held no documentation.
In 2023, cryptography engineer Filippo Valsorda and others offered a bounty for a solution, which ultimately grew to over $36,000. Despite the passing of NSA mathematician Jerry Solinas, who had provided the seeds, the mystery remained unsolved until Claude's intervention this month. The AI model successfully reproduced the sequence for P-256, a widely used curve, finding an exact match for its published seed.
Anthropic clarified that the NSA's seeds were public from their inception, meaning there was no secret data to exploit. The predictability of the lrand48 program, which allowed Claude to reconstruct the numbers, would be disastrous if used for generating private keys, as anyone could recreate them. However, in this context, the seeds were public parameters, not secret keys.
Why it matters
This resolution is significant for cybersecurity. The long-standing suspicion of an NSA backdoor in widely adopted cryptographic standards has been a source of concern for governments, corporations, and individuals relying on secure communication. The finding that the seeds originated from a common, albeit somewhat predictable, random-number generator largely alleviates these fears. It suggests the NSA's secrecy was likely due to mundane origins rather than malicious intent, though the agency never explained why certain candidates were passed over or how other parts of the curves were chosen.
Crucially, Bitcoin's security remains unaffected by this discovery. Bitcoin wallets operate on a different elliptic curve, secp256k1, which is defined by simpler, fixed numbers. This design offers fewer adjustable parameters, reducing the theoretical possibility for a hidden weakness. Satoshi Nakamoto, Bitcoin's creator, never publicly explained the choice of secp256k1, which was rarely used before Bitcoin popularized it. Ethereum later adopted it, and Bitcoin retained it even after its 2021 Taproot upgrade.
However, a separate, more distant threat to Bitcoin and other cryptocurrencies persists: quantum computing. A sufficiently powerful quantum computer could theoretically reverse-engineer a secret key from a public key visible on Bitcoin's blockchain. While hardware advancements, such as Google's recent Willow chip, demonstrate progress in reducing quantum computing errors, no quantum computer yet exists that is large or reliable enough to crack a Bitcoin key. This remains a theoretical danger without a clear timeline for realization.
What to watch
While the immediate security implications for NIST curve users are minimal—Anthropic advises no changes are needed—the episode highlights the ongoing importance of transparency in cryptographic standard setting. The $36,864 bounty for solving the mystery will be donated to the Internet Security Research Group, the nonprofit behind Let's Encrypt, which provides free security certificates.
Looking ahead, the role of artificial intelligence in solving complex, long-standing technical challenges is becoming increasingly evident. This case demonstrates AI's potential to contribute to cybersecurity by demystifying obscure data. For the broader digital asset space, continued vigilance against future threats, particularly the long-term development of quantum computing capabilities, will remain paramount.
For Information Purposes Only. Not Investment Advice. How we work