When History Lies: Evaluating and Improving Tool Use under Misleading Multi-Turn Histories

2026-08-06

AI Summary

This work addresses the challenge of historical contamination in multi-turn tool use, where agents often make decisions based on outdated or erroneous states. To tackle this issue, the authors construct a paired benchmark comprising three views—original, contaminated, and ground-truth states—and propose a reliable state policy transfer method. This approach employs soft supervision to distill a teacher policy trained on ground-truth states into a student model that observes only contaminated history. Integrating golden retention intervention analysis with sequence-level distillation, the method achieves a balanced tool-use accuracy of 87.0% on Qwen3-1.7B, substantially outperforming existing baselines. Furthermore, it demonstrates strong generalization across tasks, models, and noisy environments.

Abstract

Tool-calling agents infer task state from accumulated dialogue and tool traces. In persistent interactions, however, historical traces may remain structurally valid and semantically plausible after they cease to be authoritative for the current request. We show that such history can hijack a policy the model already possesses: on Qwen3-1.7B, pollution flips 32.1% of decisions that are correct under the original trajectory and frequently induces reuse of corrupted entities or interface conventions. We introduce bench, a paired benchmark with synchronized Original, Polluted, and Oracle State views that preserve the system policy, current tools, latest request, and gold next action. Eleven gold-preserving interventions isolate failures in decision state, entity binding, and interface execution across complete calls and non-call decisions. We further propose ours, which transfers an Oracle-conditioned teacher policy to a student observing only polluted history through soft supervision on student-generated prefixes. On Qwen3-1.7B, ours achieves 87.0% Balanced Tool-Use Accuracy, outperforming Gold-SFT (66.3%), Oracle sequence distillation (82.3%), and off-policy token distillation (85.0%). The method scales consistently: an 8B teacher raises the same compact 1.7B student to 91.9%, while an 8B student reaches 93.0%. The resulting policies further transfer to clean histories, unseen functions, independently regenerated evaluation contexts, external tool-use benchmarks, and noisy multi-hop question answering. These results establish history reliability as a distinct tool-use bottleneck and demonstrate reliable-state policy transfer as an effective and scalable solution.

Problem

• tool use
• misleading history
• multi-turn dialogue
• history reliability
• state pollution

Innovation

• misleading history
• tool-use reliability
• policy transfer
• soft supervision
• oracle-conditioned distillation