The analytical group Anthropic has presented results that force a reconsideration of approaches to auditing cryptographic systems. Their AI model, Claude, discovered hidden vulnerabilities in two encryption algorithms—HAWK and AES. This is not about breaking existing standards, but about a fundamentally new level of analysis: the machine noticed what leading global experts had missed over years of research.

HAWK's Secret Revealed in 60 Hours

HAWK is a promising digital signature system designed with the threat of quantum computers in mind. For two years, experts found no flaws in it. Claude needed only 60 hours to identify a hidden mathematical pattern. According to calculations, breaking the smallest HAWK key is now possible in 2^38 steps instead of 2^64. This reduces computational costs by approximately 67 million times. A solution exists, but it is painful: the key size will have to be doubled, which negates the algorithm's main advantage.

Attack on AES: The "Möbius Bridge"

AES is the standard that has protected the lion's share of internet traffic since 2001. Claude could not break the full version, but it analyzed a simplified variant with 7 rounds instead of 10. The AI found an unconventional method, dubbed the "Möbius Bridge." The essence of the trick is that one stage of selection turned out to be redundant. As a result, the most effective attack was accelerated by 200 to 800 times. The work took three days, and then two specialists spent nearly a month double-checking the result.

Expert Opinion from Cryptalist

This case is a warning signal for the entire industry. If before we relied on human expertise, now AI is becoming faster and more accurate in finding vulnerabilities. For cryptocurrency networks like Bitcoin and Ethereum, this is not yet a threat—they use different mathematics. But for new post-quantum standards being prepared for implementation, the speed of AI auditing could become a decisive factor. The question is whether humans will manage to adapt to this pace, or whether we will forever be chasing a machine that has already found a breach.