Web agents monitored without internal access
A new method lets you monitor AI web agents for failure risk using only observable behavior, no model internals needed.
A new method lets you monitor AI web agents for failure risk using only observable behavior, no model internals needed.