Anthropic says Claude helped uncover weaknesses in cryptographic algorithms

rickasaurus · x · 2026-07-29

Anthropic says Claude helped its researchers uncover weaknesses in cryptographic algorithms.

In the attached write-up, the model was repeatedly guided with careful prompts over three days, generated hundreds of millions of tokens, and eventually arrived at a Möbius-bridge idea that improved the attack. A few days and roughly one billion tokens later, the team had refined the method into the version described in the paper.

Related event: Anthropic's Claude Mythos Uncovers Flaws in Post-Quantum Crypto and AES(30 posts)→

Original post →

More from Research

Research channel →