16h05 ▪ 5 min read ▪ by Evans S.
Summarize this article with:
Claude just solved in less than 30 minutes a cryptographic puzzle that had been lingering for about 25 years. Anthropic’s AI model found the origin of five numbers provided by the NSA for certain elliptic curves standardized by NIST in 2000. Long suspected of possibly hiding a backdoor, they ultimately came from an old pseudo-random generator fed by timestamps from 1997 and 1998.
In Brief
- Claude found the origin of five NIST seeds in less than 30 minutes.
- They came from the lrand48 generator initialized with timestamps.
- The discovery breaks no encryption and does not directly involve Bitcoin.
Claude finds what researchers had been looking for 25 years
The case goes back to the late 1990s. The NSA provided several numbers, called seeds, used in the generation of elliptic curves later integrated into the NIST ECDSA standard.
No one knew exactly how these numbers had been chosen. This lack of explanation fed suspicions for years, especially after revelations about the NSA’s cryptographic activities. Claude had already demonstrated surprising capabilities in cybersecurity, to the point that Anthropic acknowledged several incidents involving its AI during tests.
This time, the model was not looking for a vulnerability. An Anthropic employee simply asked it to investigate the origin of the seeds, without providing the previously favored hypothesis. Many thought they came from the hash of a humorous phrase chosen by Jerry Solinas, NSA mathematician.
Claude went elsewhere. It studied the characteristics of old pseudo-random generators available in 1990s C libraries. Less than half an hour later, it had a lead.
A simple number generator behind the NSA mystery
The culprit is called lrand48. This pseudo-random generator already existed in old C libraries. Claude detected in the seeds a signature compatible with its operation, then reconstructed its internal state.
The model ultimately found the origin of the seeds for the P-192, P-224, P-256, P-384, and P-521 curves. For P-256, the program had been initialized with a Unix timestamp. It then generated successive candidates. The seed finally published corresponds to the 1,842nd candidate.
Claude also found timestamps between May 1997 and April 1998 for the various studied curves. The result mainly changes the historical interpretation of the case.
For a long time, a hypothesis held that the NSA had tested many parameters before deliberately selecting a curve with a weakness known only to the agency.
Anthropic now considers this possibility much less likely. This does not mean all questions have disappeared. Some apparently valid candidate curves were ignored for unknown reasons. The origin of several base points remains unexplained. One seed, that of B-163, still resists. So Claude has not solved everything. But the five main seeds fueling the debate now have a reproducible origin.
Bitcoin was not broken by this AI discovery
It is important to be precise: Claude did not break either ECDSA or Bitcoin. Anthropic states this explicitly. No user of the affected NIST curves needs to change their keys or systems following this discovery.
Bitcoin uses secp256k1, not P-256. The curve chosen by Satoshi Nakamoto has particularly simple parameters. It offers much less latitude for secretly selecting values to hide a weakness.
Therefore, Claude’s discovery does not affect the current security of the network. Bitcoin, however, faces another potential cryptographic problem: quantum computers. Google has just reached a new milestone with its Willow processor, without yet having a machine capable of breaking Bitcoin signatures today.
Claude’s performance mainly tells another story.
An AI explored in a few tens of minutes a lead that human cryptographers had pursued for years. It identified a statistical structure, tested several old generators, used mathematical lattice reduction techniques, then verified the result by computation.
This type of use brings AI even closer to cryptographic and blockchain infrastructures. There was even a reward at stake. A $36,864 prize had been promised if the solution was found and the money donated to a charity. It will finally go to the Internet Security Research Group, the organization behind Let’s Encrypt. Twenty-five years of research. Less than thirty minutes for Claude.
Maximize your Cointribune experience with our "Read to Earn" program! For every article you read, earn points and access exclusive rewards. Sign up now and start earning benefits.
Join the program
A
A
Evans S.
Fascinated by Bitcoin since 2017, Evariste has continuously researched the subject. While his initial interest was in trading, he now actively seeks to understand all advances centered on cryptocurrencies. As an editor, he strives to consistently deliver high-quality work that reflects the state of the sector as a whole.
DISCLAIMER
The views, thoughts, and opinions expressed in this article belong solely to the author, and should not be taken as investment advice. Do your own research before taking any investment decisions.