Drift-Bench: Diagnosing Cooperative Breakdowns in LLM Agents under Input Faults via Multi-Turn Interaction
This addresses agent safety risks for users by enabling systematic diagnosis of failures in autonomous LLM agents, though it is incremental as it builds on existing clarification research.
The paper tackles the problem of cooperative breakdowns in LLM agents due to input faults like ambiguous expressions, introducing Drift-Bench as a diagnostic benchmark to evaluate multi-turn clarification in execution environments, showing substantial performance drops with varying effectiveness across personas and fault types.
As Large Language Models transition to autonomous agents, user inputs frequently violate cooperative assumptions (e.g., implicit intent, missing parameters, false presuppositions, or ambiguous expressions), creating execution risks that text-only evaluations do not capture. Existing benchmarks typically assume well-specified instructions or restrict evaluation to text-only, single-turn clarification, and thus do not measure multi-turn disambiguation under grounded execution risk. We introduce \textbf{Drift-Bench}, the first diagnostic benchmark that evaluates agentic pragmatics under input faults through multi-turn clarification across state-oriented and service-oriented execution environments. Grounded in classical theories of communication, \textbf{Drift-Bench} provides a unified taxonomy of cooperative breakdowns and employs a persona-driven user simulator with the \textbf{Rise} evaluation protocol. Experiments show substantial performance drops under these faults, with clarification effectiveness varying across user personas and fault types. \MethodName bridges clarification research and agent safety evaluation, enabling systematic diagnosis of failures that can lead to unsafe executions.