G-CARB: Graph-Localized Conformal Agent Risk Budget for Compositional Harm
Abstract
Small language model (SLM) agents need safety controls that track consequences across tool calls with little monitoring overhead. A private read, for example, becomes a leak when a later action sends that data outside the system. We introduce CARB (Conformal Agent Risk Budget), which calibrates when to stop an agent using a ledger of harm incurred before stopping. Under exchangeable episodes, standard conformal risk control bounds this declared loss in expectation over calibration and a future episode. G-CARB selects scorer evidence along observable dependencies from private sources to outgoing actions. The ledger still covers the entire executed history, and computing the gate score requires no additional language-model inference. On AgentDojo replay with two 14B backbones, G-CARB roughly halves scorer-input records at intermediate risk budgets while improving autonomous task completion relative to full-prefix scoring; random context of the same size achieves similar gains. Controlled examples show how retaining the relevant dependency can further avoid stopping benign work.
Related
- Graph-Localized Coverage for Conformal Prediction on Inductive Graphs
- Assessing the Role of Green Premium and Localized Price Spillovers in Budget Hotels : Evidence from Delhi National Capital Region (NCR)
- Abstention Spends Its Budget on Local Label Variance More Than on Novelty: Auditing Conformal Calibration and Selective Prediction under Molecular Scaffold Shift
- Budget-Constrained Graph Augmentation for Robust Network Design via Kirchhoff Index Minimization
- Conformal non-removability of the Brownian graph
BibTeX
@misc{yu2026gcarb,
title = {{G-CARB: Graph-Localized Conformal Agent Risk Budget for Compositional Harm}},
author = {Zijun Yu and Yu Gu and Vahid Partovi Nia and Masoud Asgharian},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.05563},
eprint = {2610.05563},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.05563},
}
RIS
TY - UNPB
TI - G-CARB: Graph-Localized Conformal Agent Risk Budget for Compositional Harm
AU - Zijun Yu
AU - Yu Gu
AU - Vahid Partovi Nia
AU - Masoud Asgharian
PY - 2026
DA - 2026/10/04
PB - arXiv
DO - 10.48550/arxiv.2610.05563
UR - https://arxiv.org/abs/2610.05563
AB - Small language model (SLM) agents need safety controls that track consequences across tool calls with little monitoring overhead. A private read, for example, becomes a leak when a later action sends that data outside the system. We introduce CARB (Conformal Agent Risk Budget), which calibrates when to stop an agent using a ledger of harm incurred before stopping. Under exchangeable episodes, standard conformal risk control bounds this declared loss in expectation over calibration and a future episode. G-CARB selects scorer evidence along observable dependencies from private sources to outgoing actions. The ledger still covers the entire executed history, and computing the gate score requires no additional language-model inference. On AgentDojo replay with two 14B backbones, G-CARB roughly halves scorer-input records at intermediate risk budgets while improving autonomous task completion relative to full-prefix scoring; random context of the same size achieves similar gains. Controlled examples show how retaining the relevant dependency can further avoid stopping benign work.
ER -