A longstanding question in cryptography history has found an answer. Anthropic announced that Claude identified the process used to generate the original seed values for several widely used NIST elliptic curves, settling a matter that has drawn scrutiny for more than two decades.
The seed values matter because the standards published their numerical values without publicly explaining how they were chosen. The U.S. National Security Agency had supplied them, prompting researchers to question whether a secret selection method could conceal some mathematical advantage.
According to Anthropic's research, a model worked backward from the published values and identified an ordinary pseudorandom generation procedure. For P-256, the process required 1,842 iterations through candidate outputs to reproduce the standardized seed. The examination also covered P-192, P-224, P-384, and P-521.
Elliptic-curve signatures are foundational to account security in many applications, including passkeys and hardware-backed authentication. Understanding how the parameters were derived can build confidence in the standards without altering the mathematics underlying existing products.
Anthropic emphasized an important limitation: recovering the seed-generation history is distinct from breaking a private key, proving the curve contains a vulnerability, or confirming that all implementations are secure. The company stated there is no urgent corrective action required.
The research highlights how artificial intelligence can assist cryptographers in reconstructing forgotten design decisions, rather than defeating encryption systems. In cryptography, understanding the rationale behind numerical parameters can be as valuable as verifying that calculations function correctly.
