Claude has solved a cryptographic number theory problem that sat unresolved for 25 years, producing a result that experts in the field are describing as a real mathematical discovery rather than a routine computation. The problem traces back to research connected to the U.S. National Security Agency, and its solution, reached by Anthropic's AI model, is drawing attention from mathematicians and cryptographers who have long been familiar with the puzzle.
What Claude Actually Solved
The problem belongs to the domain of number theory, specifically a class of puzzles that underpin much of modern cryptography. For decades, researchers attempting to resolve it hit dead ends. Claude approached the problem and produced a solution path that checked out, according to those who reviewed it. The speed and method were unexpected. This follows a pattern that has started to draw serious attention: Claude cracked a 25-year cryptography problem in 30 minutes in a separate incident that also surprised specialists.
Key Facts
- The NSA-linked number theory problem had gone unsolved for approximately 25 years.
- Claude produced a solution that mathematicians confirmed as a genuine discovery.
- The result has implications for cryptographic research and number theory more broadly.
- Experts called the outcome surprising, with one quoted as saying "Wow, it's a real discovery."
Cryptographic problems of this type are not just academic exercises. They form the backbone of encryption standards used across financial systems, government communications, and consumer technology. A verified solution to a long-standing puzzle in this space has practical weight, even if immediate applications are not always obvious. Anthropic has not yet issued a formal statement detailing the technical specifics of how Claude arrived at its solution.
"Wow, it's a real discovery."Researcher quoted in CoinDesk coverage of Claude's solution
A Pattern of Scientific Breakthroughs
This is not an isolated case. Over recent months, Claude has been associated with a string of results that cross traditional disciplinary lines. Anthropic has also been building out its scientific research capabilities more broadly, with work spanning biology and medicine. The company launched an AI drug discovery program aimed at accelerating pharmaceutical research, and its biology lab has produced findings that researchers have compared to the significance of CRISPR-level advances. The cryptography result fits into that wider picture of Claude being applied to hard, open problems and returning unexpected answers.
What makes the NSA number mystery result stand out is the domain. Cryptography and number theory sit at the intersection of pure mathematics and national security, meaning scrutiny of any claimed solution is intense. The fact that reviewers reached for words like "discovery" rather than "computation" or "search result" signals that Claude did something more than retrieve a known answer. It appears to have generated a novel path through the problem.
For researchers tracking where AI capability is actually moving, results like this matter more than benchmark scores. Benchmark performance is measurable but often abstract. Solving a problem that human experts tried and failed to crack for a quarter century is concrete. It tells you something specific about what the model can do when pointed at a genuine open question. Those following the latest Claude AI news have seen this kind of result emerge with increasing frequency across different scientific fields.
The broader question now is whether this kind of performance is reproducible and how it generalizes. A single striking result can be an outlier. Several across different fields, from biology to cryptography, start to look like a capability shift. Whether Claude can be systematically applied to open problems in mathematics and cryptography, or whether these results depend heavily on problem framing and how queries are structured, remains something researchers will be watching closely in the months ahead.
