Witness reports are a key building block for reputation systems in open multi-agent systems in which agents, that are owned by a variety of stakeholders, continuously enter and leave the system. However, in such open and dynamic environments, these reports can be inaccurate because of the differing views of the reporters. Moreover, due to the conflicting interests that stem from the multiple stakeholders, some witnesses may deliberately provide false information to serve their own interests. Now, in either case, if such inaccuracy is not recognised and dealt with, it will adversely affect the function of the reputation model. To this end, this paper presents a generic method that detects inaccuracy in witness reports and updates the witness?s credibility accordingly so that less credence is placed on its future reports. Our method is empirically evaluated and is shown to help agents effectively detect inaccurate witness reports in a variety of scenarios where various degrees of inaccuracy in witness reports are introduced.