Cryptography breakthroughs how ai uncovers flaws in hawk and AES algorithms

Artificial intelligence is reshaping the landscape of cryptography. A recent study by Anthropic’s Frontier Red Team has uncovered previously unidentified vulnerabilities in two widely studied encryption algorithms: HAWK, a post-quantum signature scheme, and a reduced variant of AES.

AI’s breakthrough in post-quantum cryptography

Anthropic’s research team demonstrated that their AI model, a pre-release version of Mythos, could significantly weaken HAWK—a candidate in the NIST’s post-quantum cryptography standardization process. The breakthrough hinges on identifying a geometric symmetry in the Euclidean lattice isomorphism problem, the foundation of HAWK’s security.

By exploiting this symmetry, the team reduced the complexity of recovering the private key from 2⁶⁴ to 2³⁸ operations for the HAWK-256 variant. In practical terms, this translated to extracting the secret key in just 3 hours and 42 minutes using a standard 96-core server.

The discovery forced the HAWK team to withdraw their algorithm from the NIST selection process on July 29, 2026, effectively rendering it obsolete. While this doesn’t affect current security systems, it underscores the growing role of AI in cryptanalysis.

Cracking a reduced AES variant: theoretical progress with practical limits

The second part of the study focused on a weakened version of AES-128, restricted to 7 rounds instead of the standard 10. Anthropic’s AI developed a technique called the Möbius Bridge, an invariant fingerprinting method integrated into a meet-in-the-middle attack.

This innovation eliminated a step previously requiring 256 exhaustive searches, accelerating the attack by a factor of 200 to 800 compared to 2013 reference work. However, the vulnerability remains purely theoretical. The full AES-128 standard, with its 10 rounds, remains secure against this approach.

The findings highlight a critical distinction: while AI can accelerate theoretical cryptanalysis, current encryption standards remain robust. The ANSSI, France’s cybersecurity agency, recommends 192-bit keys for post-quantum security but notes minimal performance impact in transitioning from 128 to 192 bits.

AI’s dual role: unlocking vulnerabilities and raising new questions

The study also shed light on AI’s evolving capabilities in cryptography. The Mythos AI initially refused to engage with the reduced AES variant, asserting that no improvements were possible on such a well-established algorithm. After adjusting the prompting framework, the model generated approximately one billion tokens over several days to devise a new attack method.

The research required about 60 hours of autonomous computation, costing roughly $100,000 in API fees. The findings were later validated by two human researchers over a month to ensure mathematical rigor and eliminate potential hallucinations.

This episode reflects a broader trend: the bottleneck in cybersecurity is shifting from discovery to validation. AI is now generating vulnerabilities faster than humans can verify them, as evidenced by the surge in patches across major platforms like Firefox, Windows, and Apple systems.

Reflecting on the future of cryptography

The Anthropic team emphasized that these discoveries do not threaten current systems. However, they signal a new era where AI-assisted cryptanalysis could become a standard part of security research. The study also hints at future publications exploring vulnerabilities in other algorithms, further blurring the line between human ingenuity and machine-assisted discovery.

The research underscores the need for continuous innovation in cryptographic standards to stay ahead of AI-driven threats while leveraging its potential to strengthen security frameworks.