We propose an equilibrium theory of data-driven antitrust oversight in which regulators launch investigations on the basis of suspicious bidding patterns and cartels can adapt to the statistical screens used by regulators. We emphasize the use of asymptotically safe tests, i.e., tests that are passed with probability approaching one by competitive firms, regardless of the underlying economic environment. Our main result establishes that screening for collusion with safe tests is a robust improvement over laissez-faire. Safe tests do not create new collusive equilibria and do not hurt competitive industries. In addition, safe tests can have strict bite, including unraveling all collusive equilibria in some settings. We provide evidence that cartel adaptation to regulatory oversight is a real concern.