‘Wow, it’s a real discovery’: Claude cracks a 25-year NSA number mystery

The Myth of the Master Key

For decades, the crypto world has been haunted by a specific kind of ghost: the NSA backdoor. The theory was simple but terrifying. In the late 90s, when the National Security Agency handed over a set of prime numbers to be used as industry standards for encryption, everyone assumed they were rigged. The math community figured if the government gives you the locks, they probably kept a copy of the key.

We finally have an answer, and it didn't come from a whistleblower or a leaked memo. It came from an AI. Anthropic’s Claude model just helped researchers solve a 25-year-old mystery, proving that those suspicious numbers weren't carefully crafted traps. They were just the output of a basic, slightly lazy computer program. This matters because it changes how we view the intersection of state power and open-source code.

How Claude Broke the Silence

The numbers in question are part of the Digital Signature Standard (DSS). For years, researchers like Filippo Valsorda and others in the cryptography space have looked at these primes and asked: Why these specific ones? In cryptography, constants should be "nothing-up-my-sleeve" numbers. They should have a clear, verifiable origin so no one can bake in a mathematical vulnerability.

The NSA’s numbers were the opposite. They were just there. If the NSA had found a way to choose primes that looked random but possessed a hidden property, they could theoretically crack any communication using that standard. It turns out the truth is much more mundane. By prompted reasoning, Claude helped identify that the numbers were generated by a linear congruential generator—a very standard, old-school way for a computer to spit out a sequence of digits. The AI recognized the pattern of the digits, leading researchers to realize the NSA just used a specific seed in a common algorithm.

What This Means for the Paranoia Economy

For builders, this is a lesson in Occam’s Razor. We often spend our energy worrying about complex, coordinated conspiracies from three-letter agencies. We assume they are playing four-dimensional chess with our privacy. This discovery suggests they were actually just using the mid-90s version of a "default setting."

However, don't mistake this for a clean bill of health for the government. The fact that the NSA chose a non-transparent way to generate these numbers is still a failure of standard-setting. It created thirty years of doubt. In the startup world, transparency isn't just a moral choice; it's a technical requirement. If your users can't verify where your "random" variables come from, you have already lost their trust, regardless of whether you actually built a backdoor or not.

AI as a Forensic Tool

This isn't just a story about the NSA; it’s a story about what AI is actually good for. We spend a lot of time talking about AI writing emails or making bad art. This is a use case that actually moves the needle for builders: pattern recognition in legacy code. AI is becoming the ultimate digital archaeologist.

If you are working with legacy systems or trying to audit old smart contracts, these models are proving capable of spotting mathematical signatures that humans have overlooked for decades. Claude didn't "invent" a solution here; it recognized a fingerprint. For founders, this means the cost of auditing old infrastructure is dropping. You can now point a model at a piece of 20-year-old "black box" math and get a lead on how it was built.

The Bitcoin Reality Check

Whenever "NSA" and "Cryptography" appear in the same sentence, the Bitcoin community starts sweating. There has always been a fringe theory that the NSA created Bitcoin or that the SHA-256 hashing algorithm it uses has a similar hidden trapdoor. Let's be clear: this discovery doesn't change anything for Bitcoin.

Bitcoin’s constants are different. They are largely derived from the cube roots of prime numbers—a very public, very verifiable source. The NSA mystery that was just solved was about specific primes used in different standards. Bitcoin’s security remains intact, but so does its greatest threat: quantum computing. Solving a 25-year-old riddle about how a number was picked doesn't make that number any more resistant to a Shor’s algorithm attack in the future.

The Founder’s Takeaway

If you’re building in the privacy or infrastructure space, take two things away from this. First, the "monsters" in the room are often just incompetent or lazy, not necessarily diabolical. The NSA’s lack of transparency was a choice of convenience that looked like a conspiracy. Second, your legacy is permanent. The code you write today and the constants you choose will be audited by AIs much smarter than Claude in ten years.

We are entering an era of total transparency where "trust me" is no longer a viable business model. If you can't prove why a number exists in your code, someone—or something—will eventually find out the truth. Use "nothing-up-my-sleeve" numbers. Document your seeds. Because the ghosts of your technical debt will eventually be hunted down by an LLM.

The mystery wasn't a secret door; it was just a poorly documented hallway. In tech, those two things look identical until someone turns on the lights.

Read the original at CoinDesk →