AI keeps breaking old encryption mysteries — and is dunking on humans too (www.techradar.com)

Recent advancements in AI models have led to significant breakthroughs in code-breaking and cryptography, showcasing their enhanced capabilities. OpenAI's GPT-6 Astra was able to decode a 19th-century military cipher used by Napoleon's general, revealing troop positions that had remained a mystery for over 200 years. What would have taken human cryptographers weeks to decipher was accomplished by the AI in just six hours, highlighting its efficiency and persistence in tackling complex tasks. Additionally, Anthropic's Claude AI made strides in testing elliptical curve encryption, tracing the origins of several elliptic curve seeds generated by the NSA in under 30 minutes. While researchers had previously speculated about potential weaknesses in these seeds, Claude confirmed their security, assuring the community that no vulnerabilities existed. The model's playful dig at human cryptographers reflects a broader trend: AI is not only surpassing human performance in mathematical tasks but also engaging with the field in a way that prompts reflection on the limitations of human capabilities. These developments underline the growing importance of AI in cryptography and cybersecurity, potentially reshaping how we view and implement encryption technologies.