The cryptographic world received an unexpected call from AI. Anthropic's Claude model, over 60 hours of work, identified vulnerabilities in two encryption systems — HAWK and a simplified version of AES. This is not about breaking current standards, but about demonstrating how machine intelligence finds what leading experts missed over two years of research.

Currently, it is not our messengers or banking data that are at risk, but the future of cryptographic protection. Anthropic experts showed that AI is capable not just of analyzing, but of independently discovering mathematical patterns that lead to algorithm compromise.

What did Claude find in HAWK?

HAWK is a post-quantum digital signature designed with the threat of quantum computers in mind. For two years, experts found no flaws in it. Claude, in 60 hours, noticed a hidden pattern in its mathematical foundation. The result is shocking: cracking the smallest HAWK key now requires 2^38 steps instead of 2^64. This reduces computational costs by approximately 67 million times.

There is a solution to the problem, but it is painful — the key size needs to be doubled. This kills HAWK's main advantage: its compactness and speed.

Attack on AES: The "Möbius Bridge" method

The situation with AES is different. Claude did not break the full standard, but analyzed a simplified version — with 7 rounds instead of the standard 10. However, the method it discovered, called the "Möbius Bridge," accelerated the most effective attack by 200–800 times. The essence is that the AI found that one stage of the selection process was redundant.

Claude required three days of solo work for the entire task. Then, two Anthropic specialists spent nearly a month double-checking the result. This is a key point: humans can no longer keep up with the model's pace.

Why is this important for the crypto industry?

Bitcoin (BTC) and Ethereum (ETH) operate on a different principle, and these attacks do not affect them. But time is running. Google plans to transition to new standards by 2029, and U.S. government agencies have a quantum deadline of 2031. Now, NIST will have to evaluate a public attack on an algorithm that was still being considered for implementation.

Anthropic made it clear: not all attack results have been published. In the near future, the company will hold a special seminar with scientists. The question now facing the industry is: should data security verification be entrusted to artificial intelligence? My answer is yes. But with a caveat: we must learn to trust AI faster than it finds our mistakes.