Cryptology ePrint Archive: Report 2018/734

Random Number Generators Can Be Fooled to Behave Badly

George Teseleanu

Abstract: In this paper, we extend the work on purely mathematical Trojan horses initially presented by Young and Yung. This kind of mechanism affects the statistical properties of an infected random number generator (RNG) by making it very sensitive to input entropy. Thereby, when inputs have the correct distribution the Trojan has no effect, but when the distribution becomes biased the Trojan worsens it. Besides its obvious malicious usage, this mechanism can also be applied to devise lightweight health tests for RNGs. Currently, RNG designs are required to implement an early detection mechanism for entropy failure, and this class of Trojan horses is perfect for this job.

Category / Keywords: backdoor, random number generators, health tests

Original Publication (with minor differences): ICICS

Date: received 5 Aug 2018, last revised 13 Aug 2018

Contact author: george teseleanu at yahoo com

Available format(s): PDF | BibTeX Citation

Version: 20180815:130808 (All versions of this report)

Short URL: ia.cr/2018/734


[ Cryptology ePrint archive ]