The coset encoding scheme for the wiretap channel depends primarily on generating a random sequence of bits for every code block. The secret message indexes into the set of cosets, and a random vector selects the actual code word to be transmitted from that coset. In the literature, it is usually assumed that a statistically perfect uniform random generator is available for this purpose. This study looks at security issues arising from practical ways of implementing a random number generator, especially the ones based on Linear Feedback Shift Registers. Stream ciphers based on LFSRs are vulnerable to correlation attacks and their improved variants. This article considers the known-plaintext attack, where the random vector used in wiretap coset encoding is attacked under the assumption that the message is known to the eavesdropper a priori. © 2011 IEEE.