ALTSTEER: Selective Safety Steering for Moving Beyond Hard Refusals to Constructive Alternatives
2026-08-31 • Computation and Language
Computation and LanguageSoftware Engineering
AI summaryⓘ
The authors discuss a way to make large language models safer by controlling their responses without extra training. They introduce ALTSTEER, a method that smartly decides when to change the model’s output to avoid harmful replies and instead offer helpful and safe alternatives. This approach avoids simple refusals and works within a single generation step. Tests show ALTSTEER keeps the model useful while improving safe, constructive responses, especially for models that usually respond to bad requests with short refusals.
large language modelssafety alignmentactivation steeringinference-time controlrefusal-anchored steeringconstructive redirectionmodel utilitysafe completionALTSTEERdomain adaptation
Authors
Hoejoon Kwon, Byeonggeuk Lim, Kahyeon Kim, YoungBin Kim
Abstract
Safety alignment is essential for deploying large language models, requiring systems to prevent harmful compliance while preserving helpfulness on benign requests. Activation steering offers a training-free inference-time approach to safety control, but effective safety steering requires addressing two coupled questions: when to intervene and how generation should be shaped after intervention. However, existing safety steering methods remain limited along both dimensions, as their triggering mechanisms can be unstable across domains and refusal-oriented steering often yields rigid refusals rather than constructive safe guidance. To address these limitations, we propose ALTSTEER, an inference-time framework that couples selective intervention with refusal-anchored constructive redirection within a single inference pass. ALTSTEER uses an internal refusal-relevant signal to decide when to steer, and applies staged steering to shift generation from refusal-oriented control toward constructive alternatives. Evaluations on Llama-3.1 and Qwen2.5 show that ALTSTEER preserves benign utility while improving constructive safe-completion behavior, especially on models that otherwise tend to produce short refusals for harmful requests.