Elliptic curve digital signature algorithm: AI traces NIST curve seeds to 1990s code
Representative image: 3844328 / Pixabay
Anthropic said on October 16, 2026 that a Claude research model traced the long-unexplained seeds of NIST's ECDSA curves to a 1990s timestamp-seeded C random number generator.
What Happened
Anthropic said on Friday, October 16, 2026 that an internal research version of its Claude AI model had worked out where the seeds behind the standard curves of the elliptic curve digital signature algorithm (ECDSA) came from, a question that had gone unanswered for more than two decades. According to the company's Frontier Red Team, the seeds used to generate NIST's P-192, P-224, P-256, P-384 and P-521 curves were produced by lrand48, a random number generator in the standard C library, which had been started with Unix clock readings from the late 1990s. Anthropic stressed that the finding is not a break of the cryptography and that users do not need to change anything.
Key Facts
- Anthropic's Frontier Red Team published the research on October 16, 2026.
- The work concerns the seeds behind NIST's P-192, P-224, P-256, P-384 and P-521 elliptic curves, which are used in ECDSA.
- The seeds were traced to lrand48, a C-library random number generator, seeded with Unix timestamps.
- Reports place those timestamps between May 1997 and April 1998.
- Anthropic said several curve base points trace to FreeLIP, an arithmetic package by Arjen Lenstra, using small integer seeds; some base points remain unexplained.
- Anthropic said this is not a cryptographic break, and developers using ECDSA do not need to change their parameters.
- The company argued that being able to reproduce the seeds makes a deliberately planted weakness less plausible.
Why It Matters
ECDSA is a widely used method for creating digital signatures, which prove that a message or piece of software came from who it claims and was not altered. The NIST curves, especially P-256, are standard choices in that system. The curves were built from seed values that, according to the reports, had gone unexplained since the US National Security Agency supplied them. Unexplained numbers in a security standard invite the question of whether they were picked to hide a weakness, and that question has hung over these curves.
The new work offers a mundane answer: the seeds appear to be the output of an ordinary, weak 1990s random number generator started from the computer's clock. If that is right, the seeds were not hand-picked, which is the opposite of what a hidden backdoor would need. That is why Anthropic says the result makes deliberate tampering less plausible, while making clear that it does not change the security of the curves themselves.
The second story here is about AI. According to reports on the research, the model reasoned that the seeds might come from a weak, recognisable generator, wrote a scanner to test combinations of generators and output slices, and spotted tell-tale patterns in the seed values. A problem that had been open since the late 1990s was settled by an AI system working through hypotheses and code. It is an example of AI being used as a research assistant in a specialised technical field, not just as a chatbot, and it will add to discussion about how such models can be used in security research.
| Item | Detail |
|---|---|
| Curves covered | P-192, P-224, P-256, P-384, P-521 |
| Seed source found | lrand48 C-library generator |
| Clock readings | About May 1997 to April 1998 |
| Security impact | No break; no change needed |
| Open questions | Some base points still unexplained |
Impact
Short-term: Cryptographers will now check and try to reproduce the findings. Developers and companies using P-256 and the other NIST curves do not need to change keys or settings, according to Anthropic.
Long-term: If confirmed, the result could ease long-standing doubts about the NIST curves. It also shows how AI models can take on detailed forensic work in cryptography and computer history, which may encourage more AI-assisted research in security.
Who is affected: Cryptographers and security engineers, standards bodies such as NIST, developers of software that uses ECDSA, and the AI research community watching how models perform on hard technical problems.
Key Takeaway
Anthropic says its Claude model found that the mysterious seeds behind the NIST curves used in the elliptic curve digital signature algorithm came from a 1990s random number generator, which eases backdoor fears but does not change the curves' security.
Questions and Answers
What did Claude find about the elliptic curve digital signature algorithm?
Anthropic says an internal Claude research model traced the seeds of NIST's P-192, P-224, P-256, P-384 and P-521 curves, used in ECDSA, to lrand48, a C random number generator seeded with Unix timestamps from about May 1997 to April 1998.
Does this mean ECDSA or P-256 is broken?
No. Anthropic says it is not a cryptographic break and developers using ECDSA do not need to change their parameters.
Why did the origin of the seeds matter?
The seeds had no public explanation for how they were chosen, which led some to suspect a hidden weakness. Showing they came from an ordinary random number generator makes a deliberately planted backdoor less plausible, according to Anthropic.
Is everything about the curves now explained?
Not quite. Anthropic says several base points trace to Arjen Lenstra's FreeLIP package, but some base points remain unexplained.
Disclaimer: Prepared by the Peepals newsroom from publicly available sources with AI assistance. Information is accurate to the best of our knowledge at the time of publication and may change. Images may be representative. Not professional advice. Report an error via our contact page.
Sourced and fact-checked by the Peepals Global Editorial Team
NVIDIA RTX Spark laptops go on sale from $2,599 with Arm chip and Windows 11
Samsung Galaxy F70 India launch on October 18: Exynos 1330, 50MP OIS, under Rs 20,000
Reported, fact-checked and published by the Peepals Global Editorial Team.
