Reasoning models often determine refusal of harmful requests early in the chain-of-thought, with the opening sentence in distilled models fully controlling the outcome and patterns transferring across models from the same teacher.
This approach can lead to extortion or financial gain without the legal repercussions of distributing the material directly
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Where Do Reasoning Models Refuse?
Reasoning models often determine refusal of harmful requests early in the chain-of-thought, with the opening sentence in distilled models fully controlling the outcome and patterns transferring across models from the same teacher.