FaithSteer-BENCH: A Deployment-Aligned Stress-Testing Benchmark for Inference-Time Steering

Computer Science > Artificial Intelligence [Submitted on 18 Mar 2026] FaithSteer-BENCH: A Deployment-Aligned Stress-Testing Benchmark for Inference-Time Steering Zikang Ding, Qiying Hu, Yi Zhang, Hongji Li, Junchi Yao, Hongbo Liu, Lijie Hu Inference-time steering is widely regarded as a lightweight and parameter-free mechanism for controlling large language model (LLM) behavior, and prior work has often suggested that simple activation-level interventions can reliably induce targeted behavioral changes. However, such conclusions are typically drawn under relatively relaxed evaluation settings that overlook deployment constraints, capability trade-offs, and real-world robustness. We therefore introduce \textbf{FaithSteer-BENCH}, a stress-testing benchmark that evaluates steering methods at a fixed deployment-style operating point through three gate-wise criteria: controllability, utility preservation, and robustness. Across multiple models and representative steering approaches, we uncover several systematic failure modes that are largely obscured under standard evaluation, including illusory controllability, measurable cognitive tax on unrelated capabilities, and substantial brittleness under mild instruction-level perturbations, role prompts, encoding transformations, and data scarcity. Gate-wise benchmark results show that existing methods do not necessarily provide reliable controllability in deployment-oriented practical settings. In addition, mechanism-level diagnostics indicate that many steering methods induce prompt-conditional alignment rather than stable latent directional shifts, further explaining their fragility under stress. FaithSteer-BENCH therefore provides a unified benchmark and a clearer analytical lens for future method design, reliability evaluation, and deployment-oriented research in steering. Subjects: Artificial Intelligence (cs.AI) Cite as: arXiv:2603.18329 [cs.AI]   (or arXiv:2603.18329v1 [cs.AI] for this version)   https://doi.org/10.48550/arXiv.2603.18329 Focus to learn more Submission history From: Zikang Ding [view email] [v1] Wed, 18 Mar 2026 22:28:36 UTC (10,918 KB) Access Paper: HTML (experimental) view license Current browse context: cs.AI < prev   |   next > new | recent | 2026-03 Change to browse by: cs References & Citations NASA ADS Google Scholar Semantic Scholar Export BibTeX Citation Bookmark Bibliographic Tools Bibliographic and Citation Tools Bibliographic Explorer Toggle Bibliographic Explorer (What is the Explorer?) Connected Papers Toggle Connected Papers (What is Connected Papers?) Litmaps Toggle Litmaps (What is Litmaps?) scite.ai Toggle scite Smart Citations (What are Smart Citations?) Code, Data, Media Demos Related Papers About arXivLabs Which authors of this paper are endorsers? | Disable MathJax (What is MathJax?)