Safety
Tensor Trust
Tests whether models enforce access rules when faced with adversarial user inputs.
570,791items
3subjects
unknownlicense
safetydomain
textmodality
item-level responses released
Saturation status: No
About this data
Released game records of model access-granting behavior under human-written defense prompts and inputs.
570,791 items · 681,726 recorded attempts. Browse item pages to see all observations. Filters and means apply to the current page.
Loading observation records…