Safety
Scale MRT (Monitor Red Teaming)
Native final suspicion assessments made by monitor models on released SHADE-Arena and CUA-SHADE-Arena agent trajectories.
34,166items
633subjects
CC-BY-NC-4.0 for the ScaleAI/mrt dataset; MIT for the scaleapi/mrt source code. Source card prohibits training generative models on the trajectories.license
safetydomain
textmodality
item-level responses released
Response matrix
Loading response matrix…
lowhighUnobserved
Scale: Item-specific grading scales