This paper considers a model of probabilistic law enforcement in which a violator can undertake remediation that reduces the harm caused. In an optimal regime, violators are prompted to voluntarily remediate—or self‐police—by the promise of a reduced sanction when they do so. Self‐policing increases efficiency in two ways: (1) efficient remediation is achieved early and with certainty, rather than only when a violator is apprehended; and (2) the enforcement effort needed to deter violations is often reduced. Switching to a self‐policing enforcement regime leads optimally to less government enforcement activity and less deterrence.