Bitcoin Panicking? Mythos Cracks Post-Quantum Cryptography Algorithm in 60 Hours
Claude Mythos Preview, a new AI system from Anthropic, has made significant breakthroughs in fundamental cryptanalysis, targeting the mathematics underlying algorithms rather than just implementation bugs.
Its first target was HAWK, a candidate in the NIST post-quantum signature standardization process. Mythos identified a previously unexploited symmetry (a nontrivial automorphism) in the lattice structure of HAWK, effectively halving its effective key strength. It achieved this feat in just 60 hours, a task that had eluded two years of human expert review. Crucially, the human operator was not a lattice cryptography expert, primarily managing the project while AI agents debated and discovered the attack path.
Its second, more dramatic achievement was against a simplified 7-round version of AES-128, the world's most widely used symmetric encryption algorithm. Initially reluctant, stating the task was "impossible," Mythos was prompted to "try novel ideas." It then autonomously rewrote its own agent framework and proceeded to discover a novel attack method it named "Möbius Bridge." This technique bypasses a critical 256-guess step in previous "meet-in-the-middle" attacks, resulting in a speedup of 200-800 times.
Both findings are currently "harmless"—HAWK is not deployed, and full AES-128 remains secure. However, the process reveals a profound shift. Mythos completed the AES discovery in about a week, while human researchers spent nearly a month verifying its correctness. This highlights a new bottleneck: the rate of AI-driven discovery may soon outpace human capacity for verification. The research concludes with a pressing, unanswered question: what happens when such an AI finds a critical flaw in a widely deployed, real-world cryptosystem?
marsbit9 h fa
