This paper presents an overview of the PlurVA-LLM 2026 shared task on pluralistic value alignment across China, Indonesia, and Sri Lanka. To the best of our knowledge, this is the first shared task that aims to align a single large language model (LLM) to pluralistic values of multiple countries. We use 3 pluralistic value datasets: D2VBench (Hao et al., 2026) for China, Pancasila-Dilemmas (Supryadi et al., 2026) for Indonesia, and LKValues (Muthugala et al., 2026) for Sri Lanka. The shared task consists of two tracks: Track-1, a resource-constrained track (LLMs under 8B parameters) and Track-2, an open track without model-size constraints. We received registration from 20 teams across different countries. After finalization, we received 3 submissions for Track-1 and 2 submissions for Track-2. The final score is calculated as the average accuracy of the gold-standard answers across the three datasets. The submitted methods, including fine-tuning and prompt engineering, consistently surpasses the baselines. We believe that the results and findings from this shared task provide valuable insights and serve as a foundation for future research on pluralistic value alignment in LLMs. The dataset of this shared task is available in this link https://huggingface.co/datasets/TJUNLP/PlurVA-LLM-ST-2026.

Controlling the basic human values expressed by large language models (LLMs) has attracted growing attention in the mechanistic interpretability community. Recent work has made substantial attempts to steer such values through vector arithmetic or causal attribution over internal states. However, because values are complex social constructs, their mechanistic controllability still faces a major unresolved challenge: the auditability bottleneck, which makes it difficult to construct natural-language corpora that precisely capture value realization for mechanistic analysis. To address this challenge, we propose VALUEAUDITOR, a measurement-first framework that makes values behaviorally targetable before they are representationally steered. It operationalizes each value through grouped, theory-grounded response rubrics, derives graded supervision from same-scenario response contrasts, and performs inference-time steering through token-level sparse features that retain local cues of value expression. Experimental results show that VALUEAUDITOR significantly outperforms cutting-edge baseline methods on value steering benchmarks, while demonstrating robust control across values and scenarios with minimal trade-offs in response clarity. The code is publicly available at https://github.com/ZichenX/Steer-On-Rubrics