Paper 2025/2161

Attacks and Remedies for Randomness in AI: Cryptanalysis of PHILOX and THREEFRY

Thomas Eisenbarth, University of Lübeck

Hosein Hadipour, Ruhr University Bochum

Gregor Leander, Ruhr University Bochum

Felix Mächtle, University of Lübeck

Yevhen Perehuda, Ruhr University Bochum

Shahram Rasoolzadeh, Ruhr University Bochum

Jonas Sander, University of Lübeck

Cihangir Tezcan, Middle East Technical University

Abstract

In this work, we address the critical yet understudied question of the security of the most widely deployed pseudorandom number generators (PRNGs) in AI applications. We show that these generators are vulnerable to practical and low-cost attacks. With this in mind, we conduct an extensive survey of randomness usage in current applications to understand the efficiency requirements imposed in practice. Finally, we present a cryptographically secure and well-understood alternative, which has a negligible effect on the overall AI/ML workloads. More generally, we recommend the use of cryptographically strong PRNGs in all contexts where randomness is required, as past experience has repeatedly shown that security requirements may arise unexpectedly even in applications that appear uncritical at first.

@misc{cryptoeprint:2025/2161,
      author = {Jens Alich and Thomas Eisenbarth and Hosein Hadipour and Gregor Leander and Felix Mächtle and Yevhen Perehuda and Shahram Rasoolzadeh and Jonas Sander and Cihangir Tezcan},
      title = {Attacks and Remedies for Randomness in {AI}: Cryptanalysis of {PHILOX} and {THREEFRY}},
      howpublished = {Cryptology {ePrint} Archive, Paper 2025/2161},
      year = {2025},
      url = {https://eprint.iacr.org/2025/2161}
}