Anthropic has published the results of a study in which its AI model, Claude, discovered previously unknown vulnerabilities in two cryptographic algorithms. These are mathematical methods that protect messages, logins, and banking data from outsiders. It is important to emphasize: the current encryption systems used by millions of people are not yet at risk. However, the very fact that a machine identified what leading experts missed over years of work is a signal that cannot be ignored.
Vulnerability in HAWK: 67 million times faster
The first object of analysis is the HAWK protocol, a digital signature designed with future threats from quantum computers in mind. Over two years of research, experts found no flaws in it. Claude managed it in 60 hours. It discovered a hidden mathematical pattern that allows cracking the smallest HAWK key in just 2^38 steps instead of the expected 2^64. This means a reduction in computational cost by approximately 67 million times.
There is a solution to the problem, but it is painful: the key size will have to be doubled. As a result, HAWK's main advantage—compactness—will be lost. For the industry, this is a warning sign, showing that even the most carefully designed algorithms can have hidden defects.
Möbius Bridge attack on AES: 800 times faster
The second object is AES (Advanced Encryption Standard), the main encryption standard protecting the lion's share of internet traffic since 2001. Claude did not crack the full 10-round AES but analyzed its simplified version with 7 rounds. The AI found a non-standard attack method, named Möbius Bridge. The trick's essence: one of the key selection stages turned out to be redundant. As a result, the most effective attack was accelerated by 200–800 times.
Notably, Claude worked alone for three days. After that, two Anthropic specialists needed nearly a month to double-check its result. This clearly demonstrates how AI accelerates the analysis process where humans can no longer keep up with its pace.
What this means for cryptocurrencies and the future
Bitcoin (BTC) and Ethereum (ETH) use different cryptographic principles, so these attacks do not affect them. However, the overall trend is clear: time is running out. Google plans to transition to post-quantum standards by 2029, while U.S. government agencies have a deadline until 2031. NIST now must evaluate a public attack on an algorithm still under consideration for adoption. Anthropic indicated that not all results have been published. The company will soon hold a seminar with scientists.
My expert opinion: We are entering an era where AI becomes the main tool not only for creating but also for breaking cryptography. The question is not whether a neural network can find a vulnerability, but whether we can adapt before malicious actors do. The industry urgently needs to rethink its approach to security auditing.