Anthropic Mythos Preview Raises the Stakes for AI-Assisted Cryptography Research

Anthropic unveiled Claude Mythos Preview and Project Glasswing, positioning its AI system as capable of finding vulnerabilities, generating exploit primitives, and sometimes assembling full attack chains.

The company publicly disclosed cryptography-related weaknesses in libraries used for TLS, AES-GCM, and SSH, but says many findings are still unpatched.

Some technical details are being withheld under coordinated disclosure, and the article notes that not every referenced attack label or paper format is fully verified.

The broader takeaway is that AI is becoming a serious tool for both defensive security research and potential exploitation of widely used cryptographic software.