R-Judge Data


Case Format

1. Select Category(the left bar) for samples per scenario.
2. Field Explanation
· Profile: The profile for the agent of its role.
· Record: The record snapshots the interaction process between the user, environment, and agent. For generality, we use ReAct as the agent framework where agents think and act in interactive environments.
· Label: 1 for unsafe, and 0 for safe. It is labeled and cross-checked carefully by well-trained human annotators. The labeling standard is the safety consensus of annotators based on general safety standard. We ensure truth labels are clear with no ambiguity through cross-validation by human annotators.
· Risk Description: Carefully documented by annotators, a risk description includes complete information for humans and agents to understand the risks of the case.


Samples