Discovering Cryptographic Weaknesses with Claude (www.anthropic.com)

Researchers at Anthropic have leveraged their AI model, Claude Mythos Preview, to uncover significant vulnerabilities in cryptographic algorithms. The model discovered an improved attack on HAWK, a digital signature scheme designed for a post-quantum future, effectively reducing its key strength by half within just 60 hours of work. Additionally, Claude identified a new method to attack round-reduced versions of the Advanced Encryption Standard (AES), speeding up previous attacks by 200-800 times. Although these findings do not currently threaten deployed systems, they pose implications for the future of cryptographic security. The significance of these results lies in the demonstration of AI's potential to autonomously detect both implementation errors and inherent mathematical flaws within cryptographic schemes. Such capabilities are essential as the development of quantum computers looms, potentially compromising existing cryptographic systems. The collaboration between Anthropic researchers and Claude highlighted the model's ability to verify its findings and identify specific weaknesses, fostering a paradigm shift in cryptographic research. This breakthrough encourages further exploration of AI-assisted cryptanalysis, enhancing the resilience of future digital security standards.