Yee [Yee97] and Karjoth, et al. [KAG98] proposed schemes to protect the agent data. Various degrees of forward integrity were achieved by their results, among other benefits. A known vulnerability of their techniques is the truncation attack where two visited hosts (or one revisited host) can collude to discard the partial results collected between their respective visits. In this paper we propose several defenses against the truncation attack and the related growing-a-fake-stem (“stemming”) attack for the protection of the partial computation results of free-roaming agents. In Protocol N1, we use a co-signing technique to prevent the two-colluder truncation attack. Generalizations of N1 can further raise the threshold number of colluding hosts we can defend. Protocol N2 does not prevent truncation or stemming. It detects stemming and identifies the exact pair of colluders for prosecution. Protocol N2 mainly relies on mutual authentication techniques.