In this contribution, a cognitive radio scenario is considered, where a secondary user (SU) cooperative network is allowed to transmit concurrently with a primary user (PU), provided that the resulting average interference power at the PU receiver is kept below a certain threshold. The SU network uses a selection amplify-and-forward protocol. An optimum and a low-complexity suboptimum dynamic energy allocation strategy are proposed, that minimize the link outage probability between the source node and the destination node of the SU network under an average interference power constraint. Because the interference channels from the SU nodes to the PU are affected by fast fading, the realistic assumption is made that each SU node has access to the root mean-square value (rather than the instantaneous value) of the gain of its own interference channel to the PU. We show that both dynamic allocation strategies considerably outperform the optimum static allocation strategy, and that the low-complexity dynamic allocation performs nearly as well as the optimum dynamic allocation. Finally, we confirm that our dynamic algorithms show a significant performance improvement as compared to the optimal dynamic algorithm under a peak interference power constraint.