HAWK, a NIST Standard Candidate, Withdrawn After Fatal Flaw Found by AI Model Mythos

Based on a report by Dan Goodin in Ars Technica, a major shockwave has hit the National Institute of Standards and Technology (NIST) as it progresses with the standardization of post-quantum cryptography (PQC) algorithms.

The digital signature scheme “HAWK,” which had been under review in the third round of evaluations, was found to have a critical vulnerability by Anthropic’s security-focused AI model “Mythos.” This discovery forced the developers to voluntarily withdraw the algorithm.

HAWK was designed as a digital signature scheme resilient to future quantum computer-based attacks. It had successfully passed NIST’s second round of extensive security evaluations and was undergoing assessment in the third round. This phase is specifically aimed at uncovering vulnerabilities such as the one discovered here.

Following the results announced by Anthropic on Monday, HAWK’s developers declared the withdrawal of the algorithm on Tuesday. While NIST will continue its standardization efforts, HAWK will no longer be considered as a candidate.

Mythos attacked two cryptosystems. The first system is HAWK, a digital signature scheme designed to withstand future attacks from quantum computers. HAWK had survived two rounds of testing by NIST for evaluating the security of PQC algorithms through widespread testing. HAWK is currently in a third round of testing.

Anthropic revealed that its Mythos model successfully attacked two cryptographic systems. One was HAWK, and the other was the widely used AES encryption. However, the findings regarding HAWK were particularly severe, as they suggested the algorithm itself had been “broken.”

Practical Implications of the Attack

This discovery has sent ripples through the cryptography community, but there are several important caveats to consider. First, the findings are incremental and do not break any currently widely used cryptographic systems. The attack methods demonstrated a way to moderately reduce the computational effort needed to break the systems.

Second, the version tested was a weakened variant rather than the formally defined specification. Such “challenge instances” are provided by the authors of the specification for adversarial peer review, and it is standard practice to use such weakened versions in tests. Real-world cryptographic systems used in production are far more robust.

Third, even with the observed improvements, the mathematical foundational problems (primitives) that form the core of cryptographic systems remain secure, at least for the time being. Lastly, both attacks relied on methods that are unlikely to be viable outside controlled test environments.

The outcomes are incremental. They don’t break any of the cryptosystems anyone relies on today. They reveal methods for moderately reducing the work that would be required to defeat the systems.

Even with these limitations, it cannot be entirely ruled out that Anthropic’s claims may involve some level of marketing exaggeration. However, as a sign of significant progress in cryptographic techniques essential for privacy and security, the findings merit attention.

Impact on NIST Standardization Process

NIST’s PQC standardization process is a cornerstone of global efforts to prepare for the eventual advent of quantum computing. The ongoing third round focuses on evaluating digital signature schemes and key encapsulation mechanisms (KEMs) with post-quantum resilience. HAWK’s withdrawal introduces a degree of uncertainty into this process.

HAWK was a relatively new algorithm, relying on lattice-based mathematical problems. The vulnerability uncovered by Mythos appears to stem from specific structures inherent in these mathematical problems. This raises concerns about potential ripple effects on other PQC candidate algorithms.

Currently, Anthropic provides its Mythos model only to a select group of trusted users. The extent of this model’s general capabilities in the field of cryptanalysis remains unclear.

Capabilities and Limitations of the Mythos Model

Mythos is a security-focused AI model developed by Anthropic. This advancement highlights the potential of AI to identify mathematical structural weaknesses that might be overlooked by human cryptographers. However, whether the attack demonstrated in this case is practical against real-world cryptographic systems remains a separate issue.

It appears that HAWK’s developers, after analyzing the specific methods employed by Mythos, determined that it would be challenging to fundamentally fix the algorithm. In NIST’s standardization process, the ability to address identified vulnerabilities in proposed algorithms is a critical evaluation criterion.

Editorial Opinion

In the short term, this incident will likely cause delays in NIST’s PQC standardization process. With HAWK’s withdrawal, the evaluation of other candidate algorithms may take precedence. Additionally, the use of AI models in cryptographic research is expected to accelerate, leading to similar analyses of other PQC candidates. For the cryptography community as a whole, this serves as a pivotal moment to re-emphasize the importance of AI-driven security evaluations.

From a long-term perspective, the significance of AI’s potential to make practical cryptanalysis possible cannot be overstated. While the recent attacks were only effective on weakened instances, the emergence of more powerful AI models in the future could pose real threats to existing cryptographic systems. A potential arms race between AI and cryptographic technologies could ensue, necessitating the evolution of post-quantum cryptographic designs to anticipate AI-based attacks.

From the editorial desk’s point of view, the key question remains whether Mythos’s findings are based on a reproducible methodology rather than being a mere stroke of luck. With Anthropic yet to release detailed technical reports, third-party verification is eagerly awaited.

References

• “Mythos attack on 3rd-round PQC algorithm candidate puts it out of commission”, by Dan Goodin — Ars Technica, 2026-07-29T22:07:06.000Z (CC BY-NC-ND)
• Source URL: https://arstechnica.com/security/2026/07/mythos-uncovers-crypto-weaknesses-that-went-unknown-for-years/