Don't Lose Focus: Activation Steering via Key-Orthogonal Projections

TL;DR

This paper introduces SKOP, a novel activation steering method for LLMs that preserves attention on key tokens, improving the trade-off between steering effectiveness and utility, especially in long-context retrieval.

Contribution

SKOP constrains attention rerouting during activation steering, maintaining focus on important tokens and enhancing performance trade-offs compared to existing methods.

Findings

SKOP reduces utility degradation by 5-7x.

SKOP retains over 95% of vanilla steering efficacy.

In long-context retrieval, SKOP maintains robust performance.

Abstract

Activation steering controls LLM behaviour towards target behaviour by intervening in internal representations, yet it often degrades reasoning and retrieval performance. We argue that a primary cause of this trade-off is attention rerouting: steering vectors alter query-key matching, shifting attention away from contextually important tokens toward less informative ones. To address this, we propose Steering via Key-Orthogonal Projections (SKOP), a steering method that constrains harmful attention rerouting without eliminating steering efficacy. SKOP achieves this by preserving attention patterns on a small set of focus tokens the model relies on for reasoning and retrieval, while allowing redistribution among less critical tail tokens. Across multiple steering benchmarks, we show that SKOP achieves the best joint steering-utility trade-off, reducing utility degradation by 5-7x while retaining over 95% of vanilla steering efficacy. Our results further suggest that, in long-context retrieval settings where vanilla steering approaches are ineffective, SKOP can maintain robust performance by avoiding attention rerouting.