This paper deals with the problem of co-existence in the same environment of agents with different attitudes towards each other. In previous works of the same authors, agents endowed with different help-seeking and help-giving attitudes have been defined and simulated in a multi-agent system, and the effect of their behavior on the global performance of the system has been measured. It has been shown that the "social" agents, which both give and ask for help, are the most successful, and may be able to tolerate very well the presence of exploiters. In this paper the robustness of the social strategy is tested by putting help-giving under risky conditions. The following three variables have been manipulated: number of resources available; threshold triggering the help-giving behavior; and ratio between different kinds of exploiters and social agents. The results show that, with a few exceptions, the social strategy is quite robust in several risky conditions. Some considerations about the usefulness of such a strategy for multi-agent systems are also outlined.