Emerging from Ground: Addressing Intent Deviation in Tool-Using Agents via Deriving Real Calls into Virtual Trajectories

2026-01-21

AI Summary

Tool-calling agents often exhibit unreliable behaviors and are difficult to evaluate due to misaligned intentions. To address this, this work proposes RISE, a novel approach featuring a “real-to-synthetic” data synthesis mechanism. It first extracts verified tool primitives from real-world tool invocation logs and generates diverse negative samples via parameter perturbation to construct synthetic trajectories. A two-stage preference fine-tuning strategy is then employed to align agent intentions with user requirements. This method effectively mitigates distributional shift, achieving significant improvements of 35.28% and 23.27% on the AccTask and AccIntent metrics, respectively, and outperforming existing state-of-the-art methods across all eight evaluation benchmarks.

Technology Category

• Humans and AI: Human-Aware Planning and Behavior Prediction
• Multiagent Systems: Adversarial Agents
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language

Application Category

• Search and Retrieval-Augmented AI: Agentic search
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web

Abstract

LLMs have advanced tool-using agents for real-world applications, yet they often lead to unexpected behaviors or results. Beyond obvious failures, the subtle issue of"intent deviation"severely hinders reliable evaluation and performance improvement. Existing post-training methods generally leverage either real system samples or virtual data simulated by LLMs. However, the former is costly due to reliance on hand-crafted user requests, while the latter suffers from distribution shift from the real tools in the wild. Additionally, both methods lack negative samples tailored to intent deviation scenarios, hindering effective guidance on preference learning. We introduce RISE, a"Real-to-Virtual"method designed to mitigate intent deviation. Anchoring on verified tool primitives, RISE synthesizes virtual trajectories and generates diverse negative samples through mutation on critical parameters. With synthetic data, RISE fine-tunes backbone LLMs via the two-stage training for intent alignment. Evaluation results demonstrate that data synthesized by RISE achieve promising results in eight metrics covering user requires, execution trajectories and agent responses. Integrating with training, RISE achieves an average 35.28% improvement in Acctask (task completion) and 23.27% in Accintent (intent alignment), outperforming SOTA baselines by 1.20--42.09% and 1.17--54.93% respectively.

Problem

• intent deviation
• tool-using agents
• negative samples
• distribution shift
• preference learning

Innovation

• intent deviation
• tool-using agents
• real-to-virtual synthesis
• negative sampling
• preference learning