We propose a theory of equilibrium antitrust oversight in which: (i) regulators launch investigations on the basis of suspicious bidding patters; (ii) cartels can adapt to the statistical screens used by regulators; and may in fact use them to enforce cartel compliance. We emphasize the use of safe tests; i.e. tests that can be passed by competitive players under a broad class of environments. Such tests do not hurt competitive industries and do not help cartels support new collusive equilibria. We show that optimal collusive schemes in plausible environments fail natural safe tests; and that cartel responses to such tests explain unusual patterns in bidding data from procurement auctions held in Japan. This provides evidence that adaptive responses from cartels is a real concern that data-driven antitrust frameworks should take into account.