We consider a situation where an agent's effort is monitored by a supervisor who cares for the agent's well being. This is modeled by incorporating the agent's utility into the utility function of the supervisor. The first best solution can be implemented even if the supervisor's preferences are unknown. The corresponding optimal contract is similar to what we observe in practice: The supervisor's wage is constant and independent of his report. It induces one type of supervisor to report the agent's performance truthfully, while all others report favorably independent of performance. This implies that overstated performance (leniency bias) may be the outcome of optimal contracts under informational asymmetries.