With Andrew Clausen.

 We study a problem in which a firm can bribe an inspector to conceal evidence of illegal pollution. We find that the cheapest way to deter bribes is (i) to secretly select either the firm or the inspector to `win' a reward whenever evidence is reported; and (ii) to give both the firm and the inspector a secret clue about who will win. If the the inspector conceals the evidence, then the winner forgoes their reward --- i.e. they 'get a lemon'. The distribution of clues is carefully constructed to engineer the worst possible lemons problem in the market for concealment: player i only enters the market if her clue is strong enough to make her believe that player j is the winner, despite knowing that player j only enters if his clue indicates that player i is the winner. But then higher order reasoning leads neither players to enter the market, no matter what clue they receive. Hence, bribery never takes place in equilibrium. As well as deterring bribes cheaply and robustly, this result demonstrates the full extent of contagious adverse selection in bilateral trades.