Synergies between AI and Cryptography: Challenges and Research Directions

Abstract. The interplay between Artificial Intelligence (AI) and cryptography has been a fruitful one for many years, although it became quite prominent only recently due to the success of deep learning. In this talk, we give a broad overview of the research at the intersection of AI and cryptography, considering both directions: how to use AI techniques to design and cryptanalyze cryptographic primitives, and how cryptography can help in building secure and private AI models. For the first direction, we survey AI-based optimization methods and computational models to optimize low-level cryptographic primitives, and how deep neural networks have been successfully trained as differential distinguishers for small symmetric ciphers. In the other direction, we mention a few applications of privacy enhancing technologies applied to machine learning models. Next, we highlight some recent ideas for detecting text generated by large language models via statistical watermarking, and how to inject cryptographic backdoors in neural networks. We also take the chance to advertise the AICrypt 2024 workshop (affiliated with Eurocrypt), which is focused on these topics.

Suggested readings: https://aicrypt2024.aisylab.com;  ia.cr/2021/1092;  arXiv:2301.08012;  doi:10.1007/978-3-030-98795-4_1