Search papers, labs, and topics across Lattice.
This paper introduces Janus, a foresight-oriented framework designed to enhance long-horizon agent safety by training guards to anticipate delayed risks from partial trajectories. By employing multi-agent simulation, Janus learns a shared policy through two coupled tasks: forecasting safety-relevant futures and adjudicating safety based on both observed and anticipated data. The resulting guard model, Vanguard, significantly outperforms baseline guards, improving average protection by 15.9 percentage points while also increasing benign task completion by 5.1 percentage points across various benchmarks.
Vanguard not only blocks unsafe actions before they occur but also boosts benign task completion, achieving a remarkable dual improvement in agent safety.
Agent safety is moving from content moderation toward preventing operational failures before tool-using agents act. We propose Janus, a foresight-oriented framework for long-horizon agent safety that trains guards to anticipate delayed risks from partial trajectories. Janus synthesizes diverse agent trajectories via multi-agent simulation and learns a shared policy with two coupled tasks: an anticipation task that forecasts safety-relevant futures and an adjudication task that decides safety from both the observed prefix and anticipated future. The two tasks are jointly optimized with CoAA-RL, which rewards forecasts by their utility for downstream safety judgment. The resulting guard model, Vanguard, blocks unsafe actions before execution. Across four agent-safety benchmarks, Vanguard improves average protection by 15.9 percentage points over baseline guards while increasing benign task completion by 5.1 percentage points.
