- Anthropic’s Claude found new flaws in HAWK and a reduced version of AES.
- Anthropic says the findings do not affect today’s production systems.
- Samson Mow warns against rushing Bitcoin into post-quantum signatures.
Anthropic says its latest AI model has uncovered new weaknesses in cryptographic algorithms that had survived years of expert review. The findings do not put today’s internet users at risk, but they show that advanced AI is becoming capable of discovering flaws that human researchers previously missed.
The announcement also drew a response from JAN3 CEO Samson Mow, who said the research supports his long-standing view that Bitcoin should move cautiously when adopting post-quantum cryptography.
Claude Finds New Weaknesses in HAWK and AES
Anthropic said its Claude Mythos Preview model discovered improved attacks against two widely studied cryptographic systems.
The first involved HAWK, a post-quantum digital signature scheme currently being evaluated as part of the US National Institute of Standards and Technology (NIST) standardization process. HAWK was designed to remain secure against future quantum computers and had already passed two years of expert review.
Working largely on its own, Claude found a previously unknown mathematical symmetry in HAWK after about 60 hours of research. The discovery reduced the estimated work needed to recover the smallest HAWK key from 2^64 operations to 2^38, roughly 67 million times less computation than previously believed.
Anthropic said the practical fix would be to double HAWK’s key size. However, that change would remove many of the efficiency advantages that made HAWK an attractive post-quantum candidate.
The second result involved AES, the Advanced Encryption Standard that has protected internet traffic since 2001.
Anthropic added that Claude did not break the full AES algorithm. Instead, it found a new attack against a weakened seven-round version of AES-128, while the production version uses ten rounds. Claude developed a technique called the “Möbius Bridge,” removing one guessing step and making the best-known attack between 200 and 800 times faster.
Claude worked autonomously for three days before researchers spent nearly a month validating the result.
No Immediate Risk, But AI Is Changing Cryptography
Anthropic said neither discovery affects production systems today. HAWK is still a research proposal and has not been deployed, while the AES attack applies only to a reduced-round version used for academic research rather than the encryption protecting today’s internet.
The company said its work demonstrates that frontier AI models are beginning to contribute to cryptographic research by finding mathematical weaknesses instead of simply identifying software implementation bugs.
Samson Mow Urges Caution on Bitcoin
Following Anthropic’s announcement, Samson Mow argued that the research strengthens the case for a conservative approach to Bitcoin upgrades.
Mow recalled warning months ago against rushing Bitcoin from ECDSA and Schnorr signatures to post-quantum signature schemes before quantum computers become a real threat.
According to Mow, adopting post-quantum cryptography too early could expose Bitcoin to classical computing attacks if vulnerabilities are later discovered in newer algorithms. Mow wrote, “Make Bitcoin safe against quantum computers just to get pwned by normal computers.”
He claimed that Bitcoin should transition to post-quantum signatures only when practical quantum computers present a clear threat, rather than making preventive changes that could introduce new risks.
Mow also praised Blockstream researcher Nickler and the company’s work on hash-based signature schemes such as SHRINCS and SHRIMPS. He said these systems have a smaller mathematical attack surface than lattice-based alternatives like HAWK. SHRINCS has already been deployed on Liquid through the Simplicity smart contract framework.
Related: AI Spending Fears Push Nasdaq to 7-Week Low as Bitcoin Drops Below $63K
Disclaimer: The information presented in this article is for informational and educational purposes only. The article does not constitute financial advice or advice of any kind. Coin Edition is not responsible for any losses incurred as a result of the utilization of content, products, or services mentioned. Readers are advised to exercise caution before taking any action related to the company.
