> ## Content Index
> Fetch the complete content index at: https://wire.fourthweb.ai/llms.txt
> Use this file to discover other available public pages before exploring further.

# Claude Cracks 30-Year Cryptography Mystery By Reading Government Meeting Notes
- URL: https://wire.fourthweb.ai/claude-cracks-30-year-cryptography-mystery-by-reading-government-meeting-notes/
- Published: 2026-10-18T14:00:51.000Z
- Updated: 2026-10-18T14:00:53.000Z
- Description: The most widely used elliptic curves in cryptography were chosen using "random" seeds that nobody could explain — until Claude decided to read 10 million lines of government meeting notes.
- Author: Travis Wright
- Tags: AI Agent Economy, AI Agents, Anthropic

**The most widely used elliptic curves in cryptography were chosen using "random" seeds that nobody could explain — until** [**Claude**](https://wire.fourthweb.ai/tag/anthropic/) **decided to read 10 million lines of government meeting notes.**

### The Summary

- [Anthropic's Claude solved a 26-year cryptography mystery](https://www.anthropic.com/research/ecdsa-curve-origins?ref=wire.fourthweb.ai) by discovering the origin of unexplained seed values in NIST's P-256 and P-384 elliptic curves used across the internet
- The [AI agent](https://wire.fourthweb.ai/tag/ai-agents/) processed millions of lines of archived government meeting transcripts to trace the seeds back to specific 1997 working group decisions
- This demonstrates AI's ability to solve long-standing technical mysteries by parsing massive document sets humans couldn't economically review

### The Signal

Every time you connect to a website with HTTPS, make a crypto transaction, or verify a digital signature, you're probably using NIST's P-256 or P-384 elliptic curves. These mathematical structures secure most of the internet. But there was always a weird footnote: the "random" seeds used to generate these curves came from somewhere nobody could trace. The official NIST documentation just listed the numbers with no provenance.

This bothered cryptographers for decades. If you can't verify how a random seed was chosen, you can't fully trust that the curve wasn't intentionally weakened. The paranoid scenario: what if those seeds were selected specifically because they create a mathematical backdoor only the selector knows about?

> "The cryptography community has been asking 'where did these numbers come from' since 1999, and the answer was always 'we don't remember.'"

[Anthropic deployed Claude](https://www.anthropic.com/research/ecdsa-curve-origins?ref=wire.fourthweb.ai) to solve this. The AI agent:

- Ingested 10+ million lines of archived NIST and IEEE working group meeting notes from 1997-1999
- Cross-referenced technical discussions about curve generation with the specific seed values
- Traced the P-256 seed to a working group decision to use digits of π, and P-384 to a similar deterministic process

The human cost of this research would have been prohibitive. Security researcher Filippo Valsorda estimates it would take a person months of full-time work to read through that volume of technical documentation, and even then you might miss the relevant passages. Claude did it in hours.

### The Implication

This is what agent labor looks like at scale. Not flashy product demos, but grinding through millions of lines of unstructured data to answer questions humans literally gave up on. The P-256 mystery sat unsolved for 26 years not because it was mathematically hard, but because nobody could justify the human time to search that many documents.

Watch for more historical mysteries getting solved this way. Every field has them — unexplained decisions in standards documents, lost rationale for design choices, orphaned data sets nobody remembers collecting. The limiting factor was always search cost. That limit just dropped to near zero.

### Sources

[Hacker News Best](https://www.anthropic.com/research/ecdsa-curve-origins?ref=wire.fourthweb.ai)