We critically analyze the state of the art in research on wireless sensor network security. Assumptions about security requirements are not always consistent with the assumptions about the nature of sensor nodes. There are deficiencies in the specification of attacker models. Work on wireless sensor network security often fails to give proper definitions and justifications of what constitutes node misbehaviour. We analyze the merits and limitations of reputation-based routing protocols as a security mechanism, and observe that in wireless sensor networks there is a strong case for using application specific cross-layer optimizations and hence a diminished demand for generic security solutions.