🤖 Mythos finds mathematical weaknesses in cryptographic algorithms
Anthropic researchers, using the Claude Mythos model, have discovered mathematical vulnerabilities in the HAWK and AES-128 algorithms. In the HAWK-256 scheme, a symmetry was found that reduces the key length from 2^64 to 2^38 operations, while the "Möbius bridge" method accelerates an attack on 7-round AES-128 by 200–800 times.
🌍 This represents a shift in AI from searching for software bugs to discovering fundamental mathematical vulnerabilities, which may require a revision of NIST's Post-Quantum Cryptography (PQC) standards.
👤 AI is becoming a serious tool in cryptanalysis, creating a "verification crisis" risk where experts may not be able to keep up with verifying new model findings.
Source 1: https://www.anthropic.com/research/discovering-cryptographic-weaknesses Source 2: https://www-cdn.anthropic.com/c88771e1bf5ee8885349eed05e5484c0e5f7e02b/aes_mobius_bridge.pdf