AI uncovers weaknesses in HAWK and reduced AES encryption algorithms

Anthropic’s Frontier Red Team has published findings on two cryptographic algorithms: HAWK and a reduced version of AES. While the discoveries are concerning, they do not pose an immediate threat to current systems. However, they underscore the growing role of artificial intelligence in cryptanalysis.

Cryptographic weaknesses revealed in post-quantum algorithm HAWK

HAWK, a digital signature scheme designed to resist quantum computing attacks, has been compromised. The algorithm was a finalist in the NIST’s post-quantum cryptography standardization process, having successfully passed two rounds of expert evaluation over two years.

Researchers leveraging Anthropic’s Mythos AI model identified a geometric symmetry in HAWK’s security framework, reducing the complexity of recovering private keys by half. For the HAWK-256 variant, the attack’s complexity dropped from 2⁶⁴ to 2³⁸ operations, enabling key extraction in just 3 hours and 42 minutes on a standard 96-core server.

This breakthrough effectively nullifies HAWK’s practical use. To restore adequate security, developers would need to double key and signature sizes, undermining the algorithm’s lightweight computational advantage—a core feature that made it appealing in the first place.

As a direct consequence, the HAWK team withdrew the algorithm from the NIST selection process on July 29, 2026. While this marks a setback for post-quantum cryptography, it does not impact other candidates in the running.

AI’s growing impact on cryptanalysis

The discovery highlights AI’s accelerating role in cryptanalysis. The HAWK analysis required approximately 60 hours of autonomous computation, costing around $100,000 in API fees if conducted through Anthropic’s commercial tools. The AI model initially resisted the task, claiming no improvements were possible on a well-established algorithm. After refinements to the research framework, the model ultimately generated about a billion tokens over several days to devise a novel attack method.

AI’s involvement extends beyond HAWK. The team also explored a reduced version of AES, a widely used encryption standard. While the findings pertain exclusively to a simplified 7-round variant of AES-128 (versus the standard 10 rounds), the innovation—a technique dubbed the Möbius Bridge—accelerated existing attacks by a factor of 200 to 800. This method eliminates a previously necessary exhaustive search step, drastically improving efficiency.

Crucially, these vulnerabilities remain confined to academic research. The full AES-128 standard, with its additional three rounds, remains secure against such attacks.

Reflecting on cryptographic standards and AI’s evolving role

These discoveries do not threaten current systems, but they prompt important questions about the future of cryptographic standards. The NIST’s latest guidelines recommend 192-bit keys for post-quantum security, a change that only marginally impacts performance compared to 128-bit keys. For those seeking even greater security, 256-bit keys offer a robust alternative.

The Anthropic team’s work underscores a broader trend: AI is transforming how vulnerabilities are discovered and analyzed. While these tools accelerate cryptanalysis, they also place unprecedented pressure on human researchers to validate and verify AI-generated findings—a challenge that mirrors the rapid pace of AI-driven innovation itself.

The discoveries involving HAWK and reduced AES serve as a reminder that AI is reshaping cryptography, not just as a potential threat but as a powerful tool for uncovering weaknesses and strengthening security protocols for the future.