The French phonological rule of liaison, whereby certain underlying word-final consonants surface only when the following word starts with a vowel, sometimes creates homophony. For instance, un œuf ‘an egg’ and un neuf ‘a nine’ are both pronounced [ɛ̃.nœf]. While homophony is cross-linguistically frequent, there is evidence that it is constrained in various ways. Here, we quantify liaison-induced homophony by comparing its occurrence in real French to that in a benchmark consisting of versions of French with modified liaison consonants. We find that liaison induces more homophony in the benchmark than in real French. This is the first evidence that a phonological rule that applies across words is subject to an anti-homophony bias.