From Behavior to Mechanism: Tracing Divergent Response Modes in Frontier Language Models.

arXiv:2608.06578v2 Announce Type: replace Abstract: Frontier language models are trained with distinct data, objectives, and safety pipelines, but whether those differences produce measurably different behavior under steering pressure has not been tested. We e