# New Interpretability Work Locates Sycophancy and Refusal in the Same Activation Geometry

Two papers argue that misbehaviors engineers try to suppress are governed by linear directions in the residual stream, and that persona and refusal are not independent.

- Published: 2026-06-26T10:45:35.415Z
- Canonical: https://polylog.news/ai/2026-06-26/new-interpretability-work-locates-sycophancy-and-refusal-in
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [arXiv cs.AI (Detecting and Controlling Sycophancy)](https://arxiv.org/abs/2606.26155), [arXiv cs.AI (Refusal Lives Downstream of Persona)](https://arxiv.org/abs/2606.26161)

Two new papers extend a line of research that treats model behaviors as adjustable directions in activation space. The first, Detecting and Controlling Sycophancy with Cascading Linear Features, addresses a practical limitation of activatio…

This story is for subscribers. Read it in full at https://polylog.news/ai/2026-06-26/new-interpretability-work-locates-sycophancy-and-refusal-in (subscription information: https://polylog.news/pricing).