StepGuard: Learning Step-Level Guardrails with Scalable Supervision and Safety–Utility Balancing
Year: 2026
Type: conference-paper
Abstract: LLM-based agents can interact with external environments through tool invocation, but this capability also introduces security risks such as file modification, information leakage, and unauthorized actions.Existing guardrails often evaluate completed trajectories, leaving preexecution monitoring of step-level actions underexplored.We propose StepGuard, a steplevel guard model that can audit completed agent trajectories and check tool actions before they are executed.To train StepGuard, we introduce StepGen, an automatic data engine that generates safe and unsafe trajectories with the same context but different actions at the risky step.To further reduce over-defense and under-defense, we propose Balance-GRPO, which dynamically balances learning between safe and unsafe actions based on their observed accuracy.Experiments show that StepGuard achieves the highest average accuracy among open-weight guard models, with performance comparable to GPT-5.4.When used to guard agents on AgentDojo and AgentDyn, Step-Guard reduces mean ASR by 77.3% relative to the no-guard setting, while mean utility drops by only 2.8 points.
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W7204262910