Robustness Analysis of Agentic AI to Inconsistent and Incomplete Tool Responses
The results show that tool failures can be recognized at the return boundary, but reliable diagnosis requires combining multiple signals, and likelihood-based signals clearly capture incomplete returns and some direct inconsistencies, while action-based signals reveal how strongly a failure changes the next decision.
Jiachen Xu, T. Pedersen, Zhongming Yao et al.
· 0 citations