Anthropic has developed Claude Mythos Preview, an AI model that autonomously discovers zero-day software vulnerabilities surpassing decades of human security research. The AI found critical bugs in OpenBSD, FFmpeg, and major cryptography libraries including TLS, AES-GCM, and SSH that secure HTTPS connections and protect DeFi infrastructure.
Mythos poses particular risks to DeFi protocols worth roughly $200 billion across Ethereum and Solana. Since DeFi code is open source, the AI can rapidly catalog weaknesses at machine speed. Friction-based security measures like multisig governance, timelocks, and traditional audits offer limited protection against AI-powered attacks.
The model remains restricted to 40 software giants including Google, Apple, and Microsoft under 'Project Glasswing.' Unlike theoretical quantum computing threats to Bitcoin, Mythos represents immediate operational security risks to existing blockchain infrastructure.
