Claude Solves 25-Year NSA Cryptography Mystery in 30 Minutes
Anthropic's Claude AI model has helped solve a cryptographic mystery that puzzled researchers for approximately 25 years: the unexplained origins of numerical values used in several elliptic-curve cryptography standards supplied by the U.S. National Security Agency (NSA).
Announced by Anthropic on October 16, 2026, the discovery traces
previously unexplained cryptographic seeds to a conventional
random-number generator called lrand48, initialized
using Unix timestamps from the late 1990s.
The investigation is significant because these numbers had raised concerns that the NSA might have deliberately selected cryptographic parameters containing a hidden weakness. Claude's findings make that possibility considerably less likely, although some questions about the original selection process remain unanswered.
Quick Summary
- Anthropic's Claude research model helped solve a 25-year-old mystery surrounding NIST elliptic-curve cryptography parameters.
- The investigation traced five NIST prime-curve seeds and five
additional ANSI X9.62 seeds to the
lrand48pseudorandom number generator. - The generator was initialized using Unix timestamps from 1997 and 1998.
- Claude identified the underlying pattern in less than 30 minutes.
- The findings make the theory of an intentionally hidden NSA backdoor less plausible.
- The discovery does not represent a cryptographic break and does not require users of the affected NIST curves to take action.
- Bitcoin was not affected by this particular mystery because it uses a different elliptic curve, secp256k1.
Key Facts
| Detail | Information |
|---|---|
| Company | Anthropic |
| AI model | Claude research model |
| Discovery announced | October 16, 2026 |
| Reported by CoinDesk | October 17, 2026 |
| Mystery duration | Approximately 25 years |
| Main technology | Elliptic-curve cryptography |
| Random-number generator | lrand48 |
| Historical timestamps | 1997–1998 |
| Investigation time | Less than 30 minutes |
| Security impact | No newly discovered cryptographic break |
| Bounty | $36,864, donated to Internet Security Research Group |
What Was the 25-Year NSA Cryptography Mystery?
The mystery involved the origins of numerical values, known as seeds, used to generate parameters for several elliptic curves included in cryptographic standards.
In 2000, the National Institute of Standards and Technology (NIST) adopted a collection of elliptic curves as part of its Digital Signature Standard. The NSA had supplied the parameters, but the organization did not publicly explain how some of the original numbers were generated.
These unexplained values became a subject of discussion among cryptographers because elliptic-curve cryptography plays an important role in securing digital communications, authentication systems, software and online transactions.
The concern was that the NSA might have deliberately selected mathematical parameters that contained a hidden weakness, potentially allowing someone with knowledge of that weakness to compromise systems relying on those curves.
Researchers spent years investigating the origins of the numbers. Cryptographers Neal Koblitz and Alfred Menezes examined the controversy in their 2015 paper, A Riddle Wrapped in an Enigma .
One historical explanation suggested that the original numbers might have been generated using a hashed humorous phrase. However, that theory did not lead to a solution.
The question remained open until Anthropic's investigation in October 2026.
How Claude Solved the Mystery
Rather than continuing to search for a hidden phrase, Claude approached the problem from a different direction.
The model investigated whether the unexplained values could have originated from an older computer-based random-number generator.
1. Identifying a Random-Number Generator
Claude examined the numerical structure of the published seeds and considered whether older pseudorandom number generators could explain their patterns.
It investigated generators commonly found in historical C software
libraries, including lrand48.
The model identified a distinctive pattern in the seed values:
certain leading bits were consistently zero. This provided an
important clue because lrand48 produces 31-bit outputs
from an internal 48-bit state.
2. Reconstructing the Original Values
Claude developed code to examine the published values and test whether they matched consecutive outputs from the suspected generator.
It also investigated possible internal states and the generator's initialization process.
For some of the seeds, Claude used mathematical techniques involving the Lenstra–Lenstra–Lovász (LLL) lattice-reduction algorithm to recover the underlying generator state from truncated outputs.
The investigation identified a common pattern involving
lrand48() and the historical initialization process.
3. Finding the Historical Timestamps
The breakthrough came when Claude traced the recovered generator states back to their initialization values.
The results pointed to Unix timestamps corresponding to dates between May 1997 and April 1998.
This provided a reproducible explanation for how the unexplained cryptographic seeds could have been generated.
For example, Claude reproduced the published P-256 seed after 1,842 candidate iterations. The same approach explained the origins of the other investigated prime-curve seeds.
4. Confirming the Results
Anthropic reported that the recovered generator states reproduced the published numerical values exactly.
This was important because the discovery was not simply a pattern that looked convincing. The team could independently reproduce the values using the identified generator and initialization process.
Anthropic also published code demonstrating how selected seeds could be reconstructed.
Which Cryptographic Standards Were Investigated?
Claude's investigation identified the origins of several numerical seeds associated with NIST and ANSI elliptic curves.
The findings included five NIST prime curves:
- P-192
- P-224
- P-256
- P-384
- P-521
Anthropic also identified five additional prime-curve seeds from the ANSI X9.62 standard.
The research went further than identifying the original seeds. Claude also traced the origins of several elliptic-curve base points to an older random-number generator used in the FreeLIP mathematical software package.
However, not every parameter was explained. The origins of several other base points and the B-163 binary-curve seed remain unresolved.
Why This Discovery Matters for Cybersecurity
The discovery matters because cryptographic standards must be trustworthy.
Organizations, developers and internet infrastructure providers depend on mathematical systems that protect sensitive information and authenticate digital activity.
When the origins of important cryptographic parameters are unclear, researchers may question whether those parameters were selected transparently or whether hidden weaknesses could exist.
Claude's investigation provides a reproducible explanation for many of the previously unexplained numbers.
This makes the possibility of an intentionally planted weakness in the investigated NIST curves less plausible.
However, the discovery should not be interpreted as proof that every historical cryptographic decision was fully transparent or that every unresolved question has been answered.
The broader lesson is that reproducibility and transparency are essential when developing and maintaining security standards.
Does This Affect Bitcoin?
No. Bitcoin was not affected by this particular discovery.
Bitcoin uses the secp256k1 elliptic curve rather than the NIST P-256 curve at the center of much of the discussion surrounding the NSA mystery.
Although both are based on elliptic-curve cryptography, they have different mathematical parameters and origins.
The discovery therefore does not demonstrate that Bitcoin wallets have been compromised or that Bitcoin's cryptographic system has been weakened.
However, a separate concern remains: quantum computing.
A sufficiently powerful and reliable quantum computer could potentially use Shor's algorithm to threaten public-key cryptographic systems, including those used by Bitcoin.
That is a different security issue from the historical NSA seed mystery.
Developers and researchers continue to investigate post-quantum cryptography as a way to prepare digital infrastructure for future quantum-computing capabilities.
Advantages and Limitations of Claude's Discovery
Advantages
- Faster investigation: Claude identified the underlying explanation in less than 30 minutes.
- Mathematical reasoning: The model applied number analysis, programming and lattice-reduction techniques.
- Reproducible findings: The identified generator and historical timestamps provide a way to reconstruct the published seeds.
- Cybersecurity research: The findings help address longstanding concerns about the origins of important cryptographic parameters.
- Human research collaboration: The investigation demonstrates how AI can assist researchers with complex technical problems.
Limitations and Unanswered Questions
- Not every parameter was explained: The origins of several base points and the B-163 seed remain unknown.
- Selection criteria remain unclear: Researchers still do not know why some potentially viable candidate seeds were skipped.
- No proof of every historical decision: Reconstructing the numbers does not reveal the complete decision-making process behind the original standards.
- Independent technical scrutiny remains important: AI-generated hypotheses and code should be reviewed and reproduced by qualified researchers.
- Not a general cryptographic solution: This achievement does not mean Claude can automatically break encryption systems or discover every mathematical vulnerability.
What Happens Next?
Anthropic has published its findings, technical explanation and reproducible code, allowing other researchers to examine the results.
The $36,864 bounty associated with identifying the cryptographic seeds will be donated to the Internet Security Research Group, the nonprofit organization behind Let's Encrypt.
Further research may investigate the remaining unexplained parameters and the reasons certain candidate values were selected or skipped.
For users and organizations relying on the investigated NIST curves, Anthropic says there is no need to make changes because of this discovery.
The findings also provide an interesting example of how AI research systems could contribute to future cybersecurity investigations, mathematical analysis and software verification.
Frequently Asked Questions
What did Claude discover about the NSA?
Claude traced previously unexplained seeds used in several NIST
elliptic-curve cryptography standards to the
lrand48 random-number generator, initialized using
historical Unix timestamps.
How long did Claude take to solve the mystery?
Anthropic reported that Claude identified the origins of the investigated seeds in less than 30 minutes.
What is lrand48?
lrand48 is a pseudorandom number generator historically
available in C software libraries. It generates numerical outputs
using an internal 48-bit state.
Was the NSA hiding a cryptographic backdoor?
The discovery makes the theory of an intentionally planted backdoor in the investigated NIST curves less plausible. However, it does not establish every detail of the original selection process.
Does this discovery affect Bitcoin?
No. Bitcoin uses secp256k1, a different elliptic curve. The discovery does not reveal a weakness in Bitcoin's curve.
Does Claude's discovery mean encryption is no longer secure?
No. The investigation did not break encryption or uncover a new attack against the affected curves. It explained the origins of previously unexplained numerical parameters.
Can AI solve other complex mathematical problems?
AI models can assist with mathematical reasoning, programming, hypothesis generation and computational research. However, their findings still require verification, particularly in security-critical applications.
The Bottom Line
Claude's investigation into the 25-year-old NSA cryptography mystery is an important example of AI-assisted scientific and technical research.
By identifying the historical random-number generator and timestamps behind previously unexplained cryptographic seeds, Anthropic provided a reproducible explanation for a question that had remained open for decades.
The discovery does not mean that Claude broke encryption, exposed a Bitcoin vulnerability or proved that all cryptographic standards are free from weaknesses.
Instead, it demonstrates something more practical: advanced AI models can help researchers investigate difficult technical questions, develop new approaches and uncover patterns that might otherwise take considerably longer to identify.
As AI systems continue to improve, their role in cybersecurity and mathematical research could become increasingly important. Human expertise, independent verification and transparent research will remain essential to ensuring that these discoveries can be trusted.
Join the Discussion
Share your thoughts, questions, or topic suggestions.