arXiv CS AI
techCenter
Agents Trust Tools Too Much: Measuring Reliance on Unreliable Toolstranslating…
arXiv:2609.05587v1 Announce Type: new
Abstract: Existing evaluations of tool-using agents primarily measure whether an agent can successfully complete diverse tasks with tools. These evaluations generally assume that tools return reliable information. However, tool returns in…
Keywords#Tools#Agents#Agents Trust#Agents Trust Tools#Measuring