Anthropic Says Claude Mythos Preview Found Cryptography Implementation Weaknesses

Short summary

Anthropic reports that its Claude Mythos Preview model identified implementation weaknesses in cryptographic libraries supporting TLS, AES-GCM, and SSH, potentially enabling certificate forgery or communication decryption. The findings come from Project Glasswing, a defensive AI security research initiative, and follow coordinated vulnerability disclosure processes. One vulnerability in the Botan library was publicly disclosed and patched, while two others remained unpatched at announcement time.

• •Claude Mythos Preview found implementation flaws in TLS, AES-GCM, and SSH crypto libraries
• •Anthropic is following coordinated disclosure; one Botan vulnerability was patched, two remain open
• •Findings are about implementation errors, not breaks of the underlying cryptographic standards

Generated with AI, which can make mistakes.