ARTIS: Agentic Risk-Aware Test-Time Scaling via Iterative Simulation

TL;DR

ARTIS introduces a risk-aware, iterative simulation framework for test-time scaling in large language models, enhancing agentic decision-making reliability without environmental risk by focusing on failure modes.

Contribution

The paper presents a novel risk-aware simulation approach that improves test-time scaling for agentic LLMs, addressing limitations of naive simulators in capturing rare failure modes.

Findings

Iterative simulation significantly improves agent reliability.

Risk-aware simulation enhances performance across models and tasks.

Naive simulators struggle with rare failure modes.

Abstract

Current test-time scaling (TTS) techniques enhance large language model (LLM) performance by allocating additional computation at inference time, yet they remain insufficient for agentic settings, where actions directly interact with external environments and their effects can be irreversible and costly. We propose ARTIS, Agentic Risk-Aware Test-Time Scaling via Iterative Simulation, a framework that decouples exploration from commitment by enabling test-time exploration through simulated interactions prior to real-world execution. This design allows extending inference-time computation to improve action-level reliability and robustness without incurring environmental risk. We further show that naive LLM-based simulators struggle to capture rare but high-impact failure modes, substantially limiting their effectiveness for agentic decision making. To address this limitation, we introduce a risk-aware tool simulator that emphasizes fidelity on failure-inducing actions via targeted data generation and rebalanced training. Experiments on multi-turn and multi-step agentic benchmarks demonstrate that iterative simulation substantially improves agent reliability, and that risk-aware simulation is essential for consistently realizing these gains across models and tasks.