Claude AI Breaks Post‑Quantum Scheme, Speeds AES Attack

Introduction

Anthropic’s latest release of the Claude AI platform has turned the cryptographic community on its head by demonstrating two remarkable feats. First, Claude was able to devise a complete end‑to‑end key‑recovery attack against the HAWK‑256 post‑quantum signature scheme, a scheme that had previously been considered robust against classical cryptanalysis. Second, the same model produced a dramatic acceleration of a seven‑round AES‑128 attack, achieving speed improvements that range from two hundred to eight hundred times faster than conventional methods. These results illustrate how large language models can move beyond text generation to become active participants in advanced security research, offering new pathways for rapid vulnerability assessment and algorithmic discovery.

How Claude AI Unlocked a Hidden Symmetry in HAWK‑256

The HAWK family of lattice‑based signatures constructs its security on high‑dimensional mathematical structures that are designed to resist known attack vectors. Within this intricate lattice, however, there exists a subtle symmetry that had not been noticed in prior academic analyses. Claude’s reasoning engine parsed the underlying equations, identified the invariant property, and then mapped it onto a targeted attack pathway. By exploiting this hidden symmetry, the system was able to prune the search space dramatically, allowing a key‑recovery exploit to be constructed without resorting to brute‑force enumeration. The implementation, when run on a 96‑core server equipped with high‑performance memory, completed the full attack cycle in approximately three hours and forty‑two minutes. This timeline demonstrates that AI‑driven cryptanalysis can now compete with, and in some cases surpass, human‑led efforts that traditionally required weeks of manual effort.

• Hidden symmetry: a previously undocumented lattice invariant that reduces the effective search space.
• Automated attack pipeline: the model generated step‑by‑step instructions for key extraction, mirroring expert cryptanalyst procedures.
• Performance benchmark: ~3.7 hours on a 96‑core machine, indicating practical feasibility for real‑world security assessments.

The 7‑Round AES‑128 Breakthrough and Its Performance Gains

In a parallel investigation, Claude turned its attention to a seven‑round variant of AES‑128, a configuration that is frequently employed to stress‑test the resilience of symmetric ciphers. Traditional cryptanalysis of AES typically requires exhaustive key searches or sophisticated differential‑linear cryptanalysis that become infeasible beyond five or six rounds. Claude discovered a novel statistical redundancy within the round function that permits a shortcut to the key recovery process. By re‑ordering operations and leveraging a specialized pruning technique, the AI‑driven attack reduced the effective number of rounds needed for a successful exploit. When executed on the same 96‑core platform, the attack finished in a fraction of the time required by classic methods, delivering a performance boost that scales between two hundred and eight hundred times, depending on the precise implementation and hardware configuration. This level of acceleration transforms the economics of cryptanalysis, making it viable to evaluate new cipher designs on realistic timescales.

• Novel redundancy exploitation: a pattern in the round transformation that cuts the required work factor dramatically.
• Speed multiplier: 200‑800× faster than brute‑force or conventional differential attacks, translating to minutes instead of days.
• Practical impact: enables rapid security validation of emerging symmetric algorithms and supports faster certification cycles.

Conclusion

Anthropic’s demonstration that Claude can autonomously breach both a post‑quantum signature scheme and a multi‑round symmetric cipher underscores the transformative potential of AI in modern cryptanalysis. As these models become increasingly adept at uncovering hidden structures and accelerating exhaustive searches, they will reshape how security teams evaluate and harden cryptographic primitives. Grivyonx Cloud is positioned to incorporate these AI capabilities into its broader suite of automation and cyber intelligence services, empowering organizations to stay ahead of emerging threats while maintaining rigorous standards of accuracy and trust.