We investigate and discuss safety and privacy preserving properties of a game-theortic based coalition algorithm KCA for forming kernel stable coalitions among information agents in face of imperfect information on actual coalition values, and changing agent society. In addition, we analyze the chances of deceiving information agents to succeed in coalition negotiations using the KCA protocol. We show that a certain type of fraud which leads to an increase of individual profit can neither be prevented nor detected, but this comes at the expense of exponentially high computation costs for the deceiving agent.