Anthropics’ Claude cracks a 25-year-old NSA cryptography mystery in under 30 minutes

What happened

Anthropic’s AI model Claude solved a 25-year-old cryptography puzzle linked to the NSA’s role in setting cryptographic standards. The mystery centered on the unexplained “seeds” behind NIST’s standard curves used for digital signatures. These seeds were provided by the NSA without clarification, leading to suspicions that they might embed backdoors. Claude quickly identified that these seeds came from “lrand48,” a simple C language random number generator, debunking fears of a covert NSA backdoor. The AI cracked this in under 30 minutes—a task that eluded human experts for decades.

Why it matters

This finding changes the risk equation around cryptographic trust in widely used standards. The suspicion that NSA supplied backdoored parameters has long pressured security experts, vendors, and governments to question the security of these curves. With Claude’s solution, it becomes far less likely those standards contain hidden vulnerabilities. That can accelerate confidence in these signature algorithms, tightening security evaluation processes and reducing the need for costly alternative cryptography designs. For builders, operators, and regulators, this reduces the trust gap tied to historical NSA secrecy and simplifies decisions about which encryption standards to support or mandate.

What to watch next

Anthropic’s rapid decode marks a new frontier in using AI to audit complex technical standards and historical puzzles. Watch for more applications of large language models and AI agents to quickly vet and validate foundational tech components. This could pressure cryptography and security organizations to adopt AI tools for faster risk assessment and scenario testing. Meanwhile, cryptographers and regulators will need to evaluate how AI-driven findings affect policy and compliance around digital signature standards and other high-stakes security controls.

AI Quick Briefs Editorial Desk