Claude AI Just Cracked a Post-Quantum Test Scheme and Found a Faster 7-Round AES Attack

What changed

Claude Mythos Preview from Anthropic found a new vulnerability in HAWK-256, a post-quantum signature scheme. The AI helped derive a full end-to-end key-recovery attack. It also discovered a much faster method to attack seven-round AES-128 encryption, speeding it up by 200 to 800 times compared to previous attempts. The key breakthrough for HAWK-256 was exploiting a previously unnoticed symmetry in its underlying lattice structure. Anthropic released code showing the attack can complete in about three hours and 42 minutes on a powerful 96-core server.

Why builders should care

Post-quantum cryptography aims to protect data from future quantum computer attacks, but these results expose new risks in schemes that supposedly combat those risks today. The HAWK-256 attack weakens trust in its security, which could force projects relying on it to reconsider or accelerate a move to alternatives. The improved AES attack method shows AI can accelerate cryptanalysis for even well-studied classical systems, putting pressure on defenders to anticipate faster exploit development as AI tools mature. Builders should reassess their cryptographic choices with these evolving threats in mind.

The practical takeaway

Relying on emerging post-quantum schemes like HAWK-256 requires caution. Quick AI-driven attacks show these algorithms are not yet unbreakable, meaning sensitive systems should maintain layered defenses or wait for more thoroughly vetted standards. The speedup in attacking seven-round AES-128 signals encrypted data considered safe with partial-round AES might now be riskier. Infrastructure teams should monitor for updates on algorithm robustness and consider quantum-safe paths that have survived extensive scrutiny.

What to watch next

Key next steps are watching how cryptography communities respond—whether HAWK-256 will drop out of favor or evolve to close this symmetry flaw. Also, expect more AI-driven breakthroughs that accelerate cryptanalysis even for classical ciphers, reshaping security timelines. Builders will want updates from Anthropic and academic researchers testing related quantum-safe schemes. Finally, keep an eye on practical adoption curves for quantum-resistant encryption as this AI-driven pressure exposes weak links faster.

AI Quick Briefs Editorial Desk