Anthropic's artificial intelligence Claude has discovered critical vulnerabilities in cryptography — what does this mean for the industry?
Anthropic's AI model Claude has achieved a breakthrough in cryptanalysis, uncovering hidden vulnerabilities in two encryption systems. These are mathematical algorithms that protect our messages, passwords, and banking data from outsiders. It is important to emphasize: the current security standards used on a mass scale are not yet under threat. However, the very fact that a machine found what leading global experts missed over years of research represents a tectonic shift.
HAWK: Two Years of Searching vs. 60 Hours of AI Work
HAWK is a promising digital signature system being developed with an eye toward resistance to quantum computing. Experts spent two years analyzing it and found no flaws. Claude needed only 60 hours to notice a hidden mathematical pattern at the core of HAWK.
According to my analysis of the provided data, the discovered vulnerability dramatically reduces the scheme's security. Previously, cracking the smallest key required 2^64 steps; now this figure has dropped to 2^38. In simple terms, the computational cost of an attack has decreased by approximately 67 million times. There is a solution, but it is unpleasant: the key size must be doubled, which effectively nullifies HAWK's main advantage—its compactness.
Möbius Bridge: A New Attack on AES
The second object of study is AES (Advanced Encryption Standard), the gold standard of encryption that has protected most internet traffic since 2001. Claude did not crack the full 10-round AES. It analyzed its simplified 7-round version.
And here, the AI exceeded expectations. It found a non-standard attack method, named Möbius Bridge. The essence is that one of the key selection stages turned out to be redundant. As a result, the most efficient known attack on this version of the cipher was accelerated by 200–800 times. Notably, Claude worked alone for three days, and then two Anthropic specialists spent nearly a month double-checking its result.
Industry Conclusions: Verification Time Is Shrinking
Anthropic directly points to a key problem: humans can no longer keep up with the model's pace of work. The main delay now occurs precisely at the stage of result verification. This is reminiscent of the 2022 story when researchers cracked another post-quantum algorithm, SIKE, in an hour on a regular laptop, but here the situation is different—the AI acts as an idea generator.
My expert assessment: For the cryptocurrency industry, including Bitcoin and Ethereum, this news is not a direct threat, as they use different cryptographic principles. However, it is a wake-up call for the entire sector. The migration to post-quantum standards, which Google plans to complete by 2029 and U.S. government agencies by 2031, must accelerate. Now, NIST faces the task of evaluating a public attack on an algorithm that was still being considered for implementation. Anthropic hints that not all results have been published and is preparing a special seminar with scientists. The question is not whether AI will break cryptography, but whether we will adapt before malicious actors do.