I present the first empirical test of the strategic ambiguity hypothesis, which posits conditions under which a formal contract is deliberately incomplete. The intuition is that if part of one party's performance is unverifiable, the optimal contract might give the other party discretion over its own (verifiable) performance. I first note that testing this prediction with real-world contracts would be infeasible. I then offer an equivalent, feasible prediction by developing a simple model that extends strategic ambiguity to repeated games with enforceability constraints. Using a unique dataset of contracts from S&P 500 tech companies, I find substantial evidence of strategic ambiguity: In California, where post-employment covenants not to compete are unenforceable, employment contracts grant employers discretion over severance payments; outside California, where noncompetes are enforceable, employers have no such discretion. These results demonstrate that sophisticated parties deliberately write incomplete contracts in order to overcome limits on legal enforcement. A policy aimed at eliminating such agreements would therefore require more powerful deterrents, such as a prohibition or whistleblower scheme.