SemEval-2026 Task 9: Detecting Multilingual, Multicultural and Multievent Online Polarization

We present SemEval-2026 Task 9, a shared task on online polarization detection, covering 22 languages and comprising over 110K annotated instances. Each data instance is multi-labeled with the presence of polarization, polarization type, and polarization manifestation. Participants were asked to predict labels in three sub-tasks: (1) detecting the presence of polarization, (2) identifying the type of polarization, and (3) recognizing the polarization manifestation. The three tasks attracted over 1,000 participants worldwide and more than 10k submission on Codabench. We received final submissions from 67 teams and 73 system description papers. We report the baseline results and analyze the performance of the best-performing systems, highlighting the most common approaches and the most effective methods across different subtasks and languages. The dataset of this task is publicly available.

Over-Refusal and Representation Subspaces: A Mechanistic Analysis of Task-Conditioned Refusal in Aligned LLMs

Aligned language models that are trained to refuse harmful requests also exhibit over-refusal: they decline safe instructions that seemingly resemble harmful instructions. A natural approach is to ablate the global refusal direction, steering the hidden-state vectors away or towards the harmful-refusal examples, but this corrects over-refusal only incidentally while disrupting the broader refusal mechanism. In this work, we analyse the representational geometry of both refusal types to understand why this happens. We show that harmful-refusal directions are task-agnostic and can be captured by a single global vector, whereas over-refusal directions are task-dependent: they reside within the benign task-representation clusters, vary across tasks, and span a higher-dimensional subspace. Linear probing confirms that the two refusal types are representationally distinct from the early transformer layers. These findings provide a mechanistic explanation of why global direction ablation alone cannot address over-refusal, and establish that task-specific geometric interventions are necessary.

Can Large Language Models Make Everyone Happy?

Misalignment in Large Language Models (LLMs) refers to the failure to simultaneously satisfy safety, value, and cultural dimensions, leading to behaviors that diverge from human expectations in real-world settings where these dimensions must co-occur. Existing benchmarks, such as SAFETUNEBED (safety-centric), VALUEBENCH (value-centric), and WORLDVIEW-BENCH (culture-centric), primarily evaluate these dimensions in isolation and therefore provide limited insight into their interactions and trade-offs. More recent efforts, including MIB and INTERPRETABILITY BENCHMARK-based on mechanistic interpretability, offer valuable perspectives on model failures; however, they remain insufficient for systematically characterizing cross-dimensional trade-offs. To address these gaps, we introduce MisAlign-Profile, a unified benchmark for measuring misalignment trade-offs inspired by mechanistic profiling. First, we construct MISALIGNTRADE, an English misaligned-aligned dataset across 112 normative domains taxonomies, including 14 safety, 56 value, and 42 cultural domains. In addition to domain labels, each prompt is classified with one of three orthogonal semantic types-object, attribute, or relations misalignment-using Gemma-2-9B-it and expanded via Qwen3-30B-A3B-Instruct-2507 with SimHash-based fingerprinting to avoid deduplication. Each prompt is paired with misaligned and aligned responses through two-stage rejection sampling to ensure quality. Second, we benchmark general-purpose, fine-tuned, and open-weight LLMs on MISALIGNTRADE-revealing 12%-34% misalignment trade-offs across dimensions.

From Native Memes to Global Moderation: Cross-Cultural Evaluation of Vision-Language Models for Hateful Meme Detection

Cultural context profoundly shapes how people interpret online content, yet vision-language models (VLMs) remain predominantly trained through Western or English-centric lenses. This limits their fairness and cross-cultural robustness in tasks like hateful meme detection. We introduce a systematic evaluation framework designed to diagnose and quantify the cross-cultural robustness of state-of-the-art VLMs across multilingual meme datasets, analyzing three axes: (i) learning strategy (zero-shot vs. one-shot), (ii) prompting language (native vs. English), and (iii) translation effects on meaning and detection. Results show that the common ``translate-then-detect'' approach deteriorate performance, while culturally aligned interventions - native-language prompting and one-shot learning - significantly enhance detection. Our findings reveal systematic convergence toward Western safety norms and provide actionable strategies to mitigate such bias, guiding the design of globally robust multimodal moderation systems.

Are Aligned Large Language Models Still Misaligned?

Misalignment in Large Language Models (LLMs) arises when model behavior diverges from human expectations and fails to simultaneously satisfy safety, value, and cultural dimensions, which must co-occur in real-world settings to solve a real-world query. Existing misalignment benchmarks-such as INSECURE CODE (safety-centric), VALUEACTIONLENS (value-centric), and CULTURALHERITAGE (culture centric)-rely on evaluating misalignment along individual dimensions, preventing simultaneous evaluation. To address this gap, we introduce Mis-Align Bench, a unified benchmark for analyzing misalignment across safety, value, and cultural dimensions. First we constructs SAVACU, an English misaligned-aligned dataset of 382,424 samples spanning 112 domains (or labels), by reclassifying prompts from the LLM-PROMPT-DATASET via taxonomy into 14 safety domains, 56 value domains, and 42 cultural domains using Mistral-7B-Instruct-v0.3, and expanding low-resource domains via Llama-3.1-8B-Instruct with SimHash-based fingerprint to avoid deduplication. Furthermore, we pairs prompts with misaligned and aligned responses via two-stage rejection sampling to enforce quality. Second we benchmarks general-purpose, fine-tuned, and open-weight LLMs, enabling systematic evaluation of misalignment under three dimensions. Empirically, single-dimension models achieve high Coverage (upto 97.6%) but incur False Failure Rate >50% and lower Alignment Score (63%-66%) under joint conditions.

From Native Memes to Global Moderation: Cros-Cultural Evaluation of Vision-Language Models for Hateful Meme Detection

Cultural context profoundly shapes how people interpret online content, yet vision-language models (VLMs) remain predominantly trained through Western or English-centric lenses. This limits their fairness and cross-cultural robustness in tasks like hateful meme detection. We introduce a systematic evaluation framework designed to diagnose and quantify the cross-cultural robustness of state-of-the-art VLMs across multilingual meme datasets, analyzing three axes: (i) learning strategy (zero-shot vs. one-shot), (ii) prompting language (native vs. English), and (iii) translation effects on meaning and detection. Results show that the common ``translate-then-detect'' approach deteriorate performance, while culturally aligned interventions - native-language prompting and one-shot learning - significantly enhance detection. Our findings reveal systematic convergence toward Western safety norms and provide actionable strategies to mitigate such bias, guiding the design of globally robust multimodal moderation systems.

When the Model Said 'No Comment', We Knew Helpfulness Was Dead, Honesty Was Alive, and Safety Was Terrified

Large Language Models (LLMs) need to be in accordance with human values-being helpful, harmless, and honest (HHH)-is important for safe deployment. Existing works use Supervised Fine-Tuning (SFT) and Mixture-of-Experts (MoE) to align LLMs. However, these works face challenges in multi-objective settings, such as SFT leading to interference between conflicting objectives, while MoEs suffer from miscalibrated routing. We term this failure mode Axis Collapse, marked by (1) disjoint feature spaces causing catastrophic forgetting, and (2) unreliable inference from misrouted experts. To resolve this, we propose AlignX, a two-stage framework. Stage 1 uses prompt-injected fine-tuning to extract axis-specific task features, mitigating catastrophic forgetting. Stage 2 deploys a MoCaE module that calibrates expert routing using fractal and natural geometry, improving inference reliability. AlignX achieves significant gains on Alpaca (Helpfulness), BeaverTails (Harmlessness), and TruthfulQA (Honesty), with +171.5% win rate, +110.1% in truthfulness-informativeness, and 4.3% fewer safety violations. It also reduces latency and memory usage by over 35% compared to prior MoEs. Results across four LLMs validate its generalizability.

Do Large Language Models Reflect Demographic Pluralism in Safety?

Large Language Model (LLM) safety is inherently pluralistic, reflecting variations in moral norms, cultural expectations, and demographic contexts. Yet, existing alignment datasets such as ANTHROPIC-HH and DICES rely on demographically narrow annotator pools, overlooking variation in safety perception across communities. Demo-SafetyBench addresses this gap by modeling demographic pluralism directly at the prompt level, decoupling value framing from responses. In Stage I, prompts from DICES are reclassified into 14 safety domains (adapted from BEAVERTAILS) using Mistral 7B-Instruct-v0.3, retaining demographic metadata and expanding low-resource domains via Llama-3.1-8B-Instruct with SimHash-based deduplication, yielding 43,050 samples. In Stage II, pluralistic sensitivity is evaluated using LLMs-as-Raters-Gemma-7B, GPT-4o, and LLaMA-2-7B-under zero-shot inference. Balanced thresholds (delta = 0.5, tau = 10) achieve high reliability (ICC = 0.87) and low demographic sensitivity (DS = 0.12), confirming that pluralistic safety evaluation can be both scalable and demographically robust.

PersoDPO: Scalable Preference Optimization for Instruction-Adherent, Persona-Grounded Dialogue via Multi-LLM Evaluation

Personalization and contextual coherence are two essential components in building effective persona-grounded dialogue systems. These aspects play a crucial role in enhancing user engagement and ensuring responses are more relevant and consistent with user identity. However, recent studies indicate that open-source large language models (LLMs) continue to struggle to generate responses that are both contextually grounded and aligned with persona cues, despite exhibiting strong general conversational abilities like fluency and naturalness. We present PersoDPO, a scalable preference optimisation framework that uses supervision signals from automatic evaluations of responses generated by both closed-source and open-source LLMs to fine-tune dialogue models. The framework integrates evaluation metrics targeting coherence and personalization, along with a length-format compliance feature to promote instruction adherence. These signals are combined to automatically construct high-quality preference pairs without manual annotation, enabling a scalable and reproducible training pipeline. Experiments on the FoCus dataset show that an open-source language model fine-tuned with the PersoDPO framework consistently outperforms strong open-source baselines and a standard Direct Preference Optimization (DPO) variant across multiple evaluation dimensions.

PersoPilot: An Adaptive AI-Copilot for Transparent Contextualized Persona Classification and Personalized Response Generation

Understanding and classifying user personas is critical for delivering effective personalization. While persona information offers valuable insights, its full potential is realized only when contextualized, linking user characteristics with situational context to enable more precise and meaningful service provision. Existing systems often treat persona and context as separate inputs, limiting their ability to generate nuanced, adaptive interactions. To address this gap, we present PersoPilot, an agentic AI-Copilot that integrates persona understanding with contextual analysis to support both end users and analysts. End users interact through a transparent, explainable chat interface, where they can express preferences in natural language, request recommendations, and receive information tailored to their immediate task. On the analyst side, PersoPilot delivers a transparent, reasoning-powered labeling assistant, integrated with an active learning-driven classification process that adapts over time with new labeled data. This feedback loop enables targeted service recommendations and adaptive personalization, bridging the gap between raw persona data and actionable, context-aware insights. As an adaptable framework, PersoPilot is applicable to a broad range of service personalization scenarios.