Safety
SafeAgentBench (TRIDENT instruction decisions)
Released TRIDENT instruction-rejection observations on the SafeAgentBench task bank. Six system configurations report whether an instruction is blocked or refused; this subset measures system behavior, not embodied task-completion success.
665items
6subjects
Official task bank: MIT; third-party result export: not declaredlicense
safetydomain
agents_and_tool_usedomain
textmodality
item-level responses released
Saturation status: No
Response matrix
Loading response matrix…
Correct (1)Incorrect (0)Unobserved
Scale: 1 = correct · 0 = incorrect