#artificial intelligence
Dec 2025
Agent Tools Orchestration Leaks More: Dataset, Benchmark, and Mitigation
Results show that TOP-R requires defenses beyond prompting alone, and introduces Library-Grounded Reverse-Inference Seed Expansion (LRSE), a four-library reverse-construction pipeline, and proposes TOP-Align, an SFT+DPO method for learning safer task-completion boundaries.
Yu Qiao, Dongqi Liu, Hong-Chang Yang et al.
· arXiv.org · 12 citations
· ⚡2