SummaryRecently, cooperative communications are used to improve the communications performance in cognitive radio networks. In this paper, we present a new model for implicit cooperation between primary users and secondary users (SUs), where SUs transparently relay packets of primary transmitters (PTs) towards their primary receivers encouraging PTs to reduce their transmit power, which in turn will reduce the total interference at some intended secondary receivers (SRs). Secondary relays work as full‐duplex amplify‐and‐forward (AF) cooperative nodes. The system is modeled as a multi‐armed restless bandit problem where the optimal policy for joint relay selection and transmit power allocation is obtained in a distributed way taking both instantaneous channel states and residual energy into consideration. We solve the problem using primal‐dual priority‐index heuristic, which can reduce the online computations and implementation complexity. Simulation results show that this model can benefit both sides where the energy consumption at PTs is reduced significantly and signal to interference plus noise ratio (SINR) at the intended SUs is increased about 50% in average.