LLM agents whistleblow on staged misconduct at rates that vary by model family, decrease with task complexity and extra tools, and increase with moral nudges in the prompt.
If that is the case, watch for sentences where the model seems to really focus on implementing the task as if it is a task in the real world
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Why Do Language Model Agents Whistleblow?
LLM agents whistleblow on staged misconduct at rates that vary by model family, decrease with task complexity and extra tools, and increase with moral nudges in the prompt.