Analysing the Safety Pitfalls of Steering Vectors

MCML Authors

Yuxiao Li

→ Group Gjergji Kasneci
 Responsible Data Science

Alina Fastowski

→ Group Gjergji Kasneci
 Responsible Data Science

Efstratios Zaradoukas

→ Group Gjergji Kasneci
 Responsible Data Science

Bardh Prenkaj

Dr.

→ Group Gjergji Kasneci
 Responsible Data Science

Gjergji Kasneci

Prof. Dr.

Core PI

Responsible Data Science

Abstract

Activation steering has emerged as a powerful tool to shape LLM behaviour without the need for weight updates. While its inherent brittleness and unreliability are well-documented, its safety implications remain underexplored. In this work, we present a systematic safety audit of steering vectors obtained with Contrastive Activation Addition (CAA), a widely used steering approach, under a unified evaluation protocol. We show that steering vectors consistently influence the success rate of jailbreak attacks, with stronger amplification under simple template-based attacks. Across LLM families and sizes, steering the model in specific directions can drastically increase (by up to 57%) or decrease (by up to 50%) its attack success rate (ASR), depending on the targeted behaviour. We attribute this phenomenon to the overlap between the steering vectors and the latent subspace of refusal behaviour. Thus, we offer a mechanistic explanation for this discovery. Together, our findings reveal the previously unobserved origin of this safety gap in LLMs, highlighting a trade-off between controllability and safety.

inproceedings LFZ+26

---

Findings @ACL 2026

Findings at the 64th Annual Meeting of the Association for Computational Linguistics. San Diego, CA, USA, Jul 02-07, 2026.

---

Authors

Y. Li • A. Fastowski • E. Zaradoukas • B. Prenkaj • G. Kasneci

---

Links

DOI GitHub

---

Research Area

 A1 | Statistical Foundations & Explainability