ARTIS: Agentic Risk-Aware Test-Time Scaling via Iterative Simulation
This work addresses the problem of costly and irreversible failures in agentic LLM interactions for AI safety and deployment, representing an incremental improvement over existing test-time scaling methods.
The paper tackles the problem of insufficient test-time scaling for agentic LLMs by proposing ARTIS, a framework that uses iterative simulation to explore actions before real-world execution, improving agent reliability without environmental risk. Experiments show that iterative simulation substantially enhances agent reliability, and risk-aware simulation is essential for consistent gains across models and tasks.
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.