What happens if the second-best outcome is not collusion-proof implementable? What will the optimal WCP auction look like in such a case? While a complete answer to this latter question is unavailable, we provide two observations that will help answer that question. First, we show that any optimal WCP auction in a general monotonic class involves “exclusion” — a positive probability that the object is not sold to any collusive bidder. Second, we characterize the optimal WCP auction completely in the linear class for the case of symmetric bidders.