Verify Before You Commit: Towards Faithful Reasoning in LLM Agents via Self-Auditing
This addresses a critical issue for developers of long-horizon AI agents by preventing the propagation of unsupported beliefs, though it is an incremental improvement over existing consensus-based methods.
The paper tackles the problem of unfaithful reasoning in LLM agents, where coherent but logically flawed intermediate beliefs lead to systematic behavioral drift, and proposes the SAVeR framework that enforces verification before action commitment, resulting in improved reasoning faithfulness across six benchmark datasets while maintaining competitive end-task performance.
In large language model (LLM) agents, reasoning trajectories are treated as reliable internal beliefs for guiding actions and updating memory. However, coherent reasoning can still violate logical or evidential constraints, allowing unsupported beliefs repeatedly stored and propagated across decision steps, leading to systematic behavioral drift in long-horizon agentic systems. Most existing strategies rely on the consensus mechanism, conflating agreement with faithfulness. In this paper, inspired by the vulnerability of unfaithful intermediate reasoning trajectories, we propose \textbf{S}elf-\textbf{A}udited \textbf{Ve}rified \textbf{R}easoning (\textsc{SAVeR}), a novel framework that enforces verification over internal belief states within the agent before action commitment, achieving faithful reasoning. Concretely, we structurally generate persona-based diverse candidate beliefs for selection under a faithfulness-relevant structure space. To achieve reasoning faithfulness, we perform adversarial auditing to localize violations and repair through constraint-guided minimal interventions under verifiable acceptance criteria. Extensive experiments on six benchmark datasets demonstrate that our approach consistently improves reasoning faithfulness while preserving competitive end-task performance.