Can you train AI to invert AES? (www.juanslozano.com)

A new exploration has emerged regarding the possibility of using AI to invert the Advanced Encryption Standard (AES), a critical algorithm underlying various security protocols like HTTPS and WPA2/3. The author proposes that, while solving the inversion problem seems formidable due to sparsity of rewards and complexity theory constraints, advancements in AI and machine learning might offer pathways to approach this challenge. The paper outlines a method for encoding AES as a satisfiability problem, specifically targeting the AES-XTS mode for full-disk encryption, to allow AI models to parse the algorithm systematically. This discussion is significant for the AI/ML community because successful inversion of AES could disrupt encryption standards vital for data security worldwide. The author delves into the mathematical underpinnings of AES, detailing its substitution-permutation structure and how AI could potentially navigate the algorithm's combinatorial nature. By leveraging techniques like lowered encoding to provide intermediate rewards for AI models, the groundwork is laid for future research. Should effective AI strategies for AES inversion be discovered, it could prompt a re-evaluation of encryption methods and spark a significant paradigm shift in cybersecurity practices.