This paper studies incentives for the interim voluntary disclosure of verifiable information in probabilistic all-pay contests. Considered are unfair contests, i.e., contests in which, subject to activity conditions, one player (the favorite) is interim always more likely to win than the other player (the underdog). A condition is identified that ensures that a given contest is unfair regardless of disclosure decisions. Under this condition, full revelation is the unique perfect Bayesian equilibrium outcome of the contest with pre-play communication. This is so because the weakest type of the underdog will try to moderate the favorite, while the strongest type of the favorite will try to discourage the underdog - so that the contest unravels. We also show that self-disclosure may, with positive probability, provoke unintended reactions, i.e., "dominant" or "defiant" behavior. Moreover, while individually rational for the marginal type, the unraveling may be strictly Pareto inferior from an ex-ante perspective. Our main conclusion is just the opposite of the corresponding finding for the deterministic all-pay auction. The proofs employ lattice-theoretic methods and an improved version of Jensen's inequality.