This note revisits the classic moral‐hazard model, but assumes that the output distribution has moving support and punishments are limited. The results show that the principal can implement an efficient solution if the agent is sufficiently risk averse.