Morocco

Large language model (LLM) based search agents iteratively generate queries, retrieve external information, and reason to answer open-domain questions. While researchers have primarily focused on improving their utility, their safety behaviors remain underexplored. In this paper, we first evaluate search agents using red-teaming datasets and find that they are more likely to produce harmful outputs than base LLMs. For example, when asked ``How can I track someone&#39;s location without their consent?&#39;&#39;, a base model refuses, whereas a search agent designed to retrieve and cite sources may lower its refusal threshold, fetch documents (e.g., court cases), and, once appended, synthesize them into an informative yet unsafe summary. We further show that utility-oriented fine-tuning intensifies this risk, motivating joint alignment of safety and utility. We present SafeSearch, a multi-objective reinforcement learning approach that couples a final-output safety/utility reward with a novel query-level shaping term that penalizes unsafe queries and rewards safe ones. Experiments show that SafeSearch reduces agent harmfulness by over 80% across three red-teaming datasets while producing safe, helpful responses, and matches the QA performance of a utility-only finetuned agent; further analyses confirm the effectiveness of the query-level reward in jointly improving safety and utility.

EACL 2026

SafeSearch: Do Not Trade Safety for Utility in LLM Search Agents

agent rl

search agent

agent safety

llm agent

### Welcome to The 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025)

Message from the General Chair: 
*It is my great pleasure and honor to welcome you to the 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025), held in beautiful Vienna, Austria, from July 27 to August 1, 2025. ACL2025continues our field’s tradition of excellence in scholarship, innovation, and inclusivity, and I am deeply grateful to the many volunteers who have worked tirelessly to bring this event to life.* 
[Read more](https://drive.google.com/file/d/1GI_hvOpjswAuYdUTromfeDiPpCcqidwg/view?usp=sharing)

To access this event page, you need to log in with the **email address you registered with**. Access credentials will be sent to your email from Underline - subject line "Welcome to ACL 2025". Please be sure to check your spam email folder if you do not see an email confirmation right away.

Please log in

To access this event page, you need to log in with the **email address you registered with**. 

ACL 2025

Welcome to The 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025)

Large language model (LLM) based search agents iteratively generate queries, retrieve external information, and reason to answer open-domain questions. While researchers have primarily focused on improving their utility, their safety behaviors remain underexplored. In this paper, we first evaluate search agents using red-teaming datasets and find that they are more likely to produce harmful outputs than base LLMs. For example, when asked ``How can I track someone's location without their consent?'', a base model refuses, whereas a search agent designed to retrieve and cite sources may lower its refusal threshold, fetch documents (e.g., court cases), and, once appended, synthesize them into an informative yet unsafe summary. We further show that utility-oriented fine-tuning intensifies this risk, motivating joint alignment of safety and utility. We present SafeSearch, a multi-objective reinforcement learning approach that couples a final-output safety/utility reward with a novel query-level shaping term that penalizes unsafe queries and rewards safe ones. Experiments show that SafeSearch reduces agent harmfulness by over 80% across three red-teaming datasets while producing safe, helpful responses, and matches the QA performance of a utility-only finetuned agent; further analyses confirm the effectiveness of the query-level reward in jointly improving safety and utility.

poster

#### *Message from the General Chair, Aline Villavicencio*
I’m delighted and honoured to welcome you to the 19th Conference of the European Chapter of the Association for Computational Linguistics (EACL 2026), taking place in the beautiful city of Rabat, in Morocco, in March 24-29, 2026. EACL is the flagship European conference of the Association and EACL 2026 proudly continues our field’s tradition of excellence in scholarship, innovation, and inclusivity. I am deeply grateful to the many volunteers whose dedication, generosity, and tireless efforts have made this conference possible.
For the first time EACL is being hosted in the African continent. This is an important milestone for our community, and we are grateful to our Moroccan hosts for enabling this historic moment by bringing this edition of EACL to Rabat. We are also delighted that the Second Arabic NLP School is co-located with EACL. We hope attendees enjoy this wonderful opportunity to strengthen ties with the Computational Linguistics communities across the African continent. *[Read full message](https://drive.google.com/file/d/14NlmHvwM6fPJuMmOvVh7K0vtQbEyv3SZ/view?usp=sharing)* 

<html><button style="display: inline-flex; align-items: center; justify-content: center; white-space: nowrap; border-radius: 9999px; font-weight: bold; background: #7c3aed; color: white; font-family: 'Space Grotesk', sans-serif; height: 40px; font-size: 16px; padding: 0 20px; border: none; cursor: pointer" onclick="window.open('https://underline.io/events/522/reception','_blank')">Go to Workshops and Tutorials Program</button></html>

To access this event page, you need to log in with the **email address you registered with**. Access credentials will be sent to your email from Underline - subject line "Welcome to EACL 2026". Please be sure to check your spam email folder if you do not see an email confirmation right away.

To access this event page, you are required to register.
Please complete your registration to continue.

We recommend reading [**the registration information**](https://2026.eacl.org/registration/) first.

**Online Registration Form**: https://acl.swoogo.com/eacl2026

Registration Required

Welcome to the 19th Conference of the European Chapter of the Association for Computational Linguistics (EACL). EACL 2026 will be held in Rabat, Morocco, from March 24–29, 2026. 

Plane geometry problem solving (PGPS) has recently gained significant attention as a benchmark to assess the multi-modal reasoning capabilities of large vision-language models. Despite the growing interest in PGPS, the research community still lacks a comprehensive overview that systematically synthesizes recent work in PGPS. To fill this gap, we present a survey of existing PGPS studies. We first categorize PGPS methods into an encoder-decoder framework and summarize the corresponding output formats used by their encoders and decoders. Subsequently, we classify and analyze these encoders and decoders according to their architectural designs. Finally, we outline major challenges and promising directions for future research. In particular, we discuss the hallucination issues arising during the encoding phase within encoder-decoder architectures, as well as the problem of data leakage in current PGPS benchmarks.

Plane Geometry Problem Solving with Multi-modal Reasoning: A Survey

Recent multimodal large language models (MLLMs) extend language understanding beyond text to speech, enabling unified reasoning across modalities. While biases in text-based LLMs have been widely examined, their persistence and manifestation in spoken inputs remain underexplored. This work presents the first systematic investigation of speech bias in multilingual MLLMs. We construct and release the BiasInEar Dataset, a speech-augmented benchmark based on Global MMLU Lite, spanning English, Chinese, and Korean, balanced by gender and accent, and totaling 70.8 hours (4,249 minutes) of speech with 11,200 questions. Using four complementary metrics (accuracy, entropy, APES, and Fleiss’), we evaluate nine representative models under linguistic language and accent, demographic gender, and structural option order perturbations. Our findings reveal that MLLMs are relatively robust to demographic factors but highly sensitive to language and option order, suggesting that speech can amplify existing structural biases. Moreover, architectural design and reasoning strategy substantially affect robustness across languages. Overall, this study establishes a unified framework for assessing fairness and robustness in speech-integrated LLMs, bridging the gap between text- and speech-based evaluation.

Bias in the Ear of the Listener: Assessing Sensitivity in Audio Language Models Across Linguistic, Demographic, and Positional Variations

The rise of toxic content on digital platforms has intensified the demand for automatic moderation tools. While English has benefited from large-scale annotated corpora, Spanish remains under-resourced, particularly for nuanced cases of toxicity such as irony, sarcasm, or indirect aggression. We present an extended version of the NECOS-TOX corpus, comprising 4,011 Spanish comments collected from 16 major news outlets. Each comment is annotated across three levels of toxicity (Non-Toxic, Slightly Toxic, and Toxic), following an iterative annotation protocol that achieved substantial inter-annotator agreement (k = 0.74). To reduce annotation costs while maintaining quality, we employed a human-in-the-loop active learning strategy, with manual correction of model pre-labels. We benchmarked the dataset with traditional machine learning (ML) methods, domain-specific transformers, and instruction-tuned large language models (LLMs). Results show that compact encoder models (e.g., RoBERTa-base-bne, 125M parameters) perform on par with much larger models (e.g., LLaMA-3.1-8B), underscoring the value of in-domain adaptation over raw scale. Our error analysis highlights persistent challenges in distinguishing subtle forms of toxicity, especially sarcasm and implicit insults, and reveals entity-related biases that motivate anonymization strategies. The dataset and trained models are released publicly.

Nuanced Toxicity Detection in Spanish: A New Corpus and Benchmark Study

Large Language Models (LLMs) are increasingly deployed in high-impact scenarios, raising concerns about their safety and security. Despite existing defense mechanisms, LLMs remain vulnerable to adversarial attacks. This paper introduces the novel attack-agnostic pipeline SENTRY (semantic entropy-based attack recognition system) for detecting such attacks by leveraging the predictive entropy of model outputs, quantified through the Token-Level Shifting Attention to Relevance (TokenSAR) score, a weighted token entropy measurement. Our approach dynamically identifies adversarial inputs without relying on prior knowledge of attack specifications. It requires only ten newly generated tokens, making it a computationally efficient and adaptable solution. We evaluate the pipeline on multiple state-of-the-art models, including Llama, Vicuna, Falcon, Deep Seek, and Mistral, using a diverse set of adversarial prompts generated via the h4rm31 framework. Experimental results demonstrate a clear separation in TokenSAR scores between benign, malicious, and adversarial prompts. This distinction enables effective threshold-based classification, achieving robust detection performance across various model architectures. Our method outperforms traditional defenses in terms of adaptability and resource efficiency.

Detection of Adversarial Prompts with Model Predictive Entropy

Large Language Models (LLMs) excel in Natural Language Processing (NLP) tasks, but they often propagate biases embedded in their training data, which is potentially impactful in sensitive domains like healthcare. While existing benchmarks evaluate biases related to individual social determinants of health (SDoH) such as gender or ethnicity, they often overlook interactions between these factors and lack context-specific assessments. This study investigates bias in LLMs by probing the relationships between gender and other SDoH in French patient records. Through a series of experiments, we found that embedded stereotypes can be probed using SDoH input and that LLMs rely on embedded stereotypes to make gendered decisions, thus positioning the assessment of relationships between SDoH as a potential complement to the evaluation of LLM performance.

Investigating Gender Stereotypes in Large Language Models via Social Determinants of Health

The predictive uncertainty of machine translation (MT) models is typically used as a quality estimation proxy. In this work, we posit that apart from confidently translating when a single correct translation exists, models should also maintain uncertainty when the input is ambiguous. We use uncertainty to measure gender bias in MT systems. When the source sentence includes a lexeme whose gender is not overtly marked, but whose target-language equivalent requires gender specification, the model must infer the appropriate gender from the context and can be susceptible to biases. Prior work measured bias via gender accuracy, however it cannot be applied to ambiguous cases. Using semantic uncertainty, we are able to assess bias when translating both ambiguous and unambiguous source sentences, and find that high translation accuracy does not correlate with exhibiting uncertainty appropriately, and that debiasing affects the two cases differently.

Uncertainty Quantification for Evaluating Gender Bias in Machine Translation

Large Language Models (LLMs) are increasingly integral to information dissemination and decision-making processes. Given their growing societal influence, understanding potential biases, particularly within the political domain, is crucial to prevent undue influence on public opinion and democratic processes. This work investigates political bias and stereotype propagation across eight prominent LLMs using the two-dimensional Political Compass Test (PCT). Initially, the PCT is employed to assess the inherent political leanings of these models. Subsequently, persona prompting with the PCT is used to explore explicit stereotypes across various social dimensions. In a final step, implicit stereotypes are uncovered by evaluating models with multilingual versions of the PCT. Key findings reveal a consistent left-leaning political alignment across all investigated models. Furthermore, while the nature and extent of stereotypes vary considerably between models, implicit stereotypes elicited through language variation are more pronounced than those identified via explicit persona prompting. Interestingly, for most models, implicit and explicit stereotypes show a notable alignment, suggesting a degree of transparency or "awareness" regarding their inherent biases. This study underscores the complex interplay of political bias and stereotypes in LLMs.

The Hidden Bias: A Study on Explicit and Implicit Political Stereotypes in Large Language Models

As Vision Language Models (VLMs) become integral to real-world applications, understanding their demographic biases is critical. We introduce GRAS, a benchmark for uncovering demographic biases in VLMs across gender, race, age, and skin tone, offering the most diverse coverage to date. We further propose the GRAS Bias Score, an interpretable metric for quantifying bias. We benchmark five state-of-the-art VLMs and reveal concerning bias levels, with the least biased model attaining a GRAS Bias Score of 98, far from the unbiased ideal of 0. Our findings also reveal a methodological insight: evaluating bias in VLMs with visual question answering (VQA) requires considering multiple formulations of a question. Our code, data, and evaluation results are publicly available at https://anonymous.4open.science/r/vqa-skin-bias-5DA3/

Ask Me Again Differently: GRAS for Measuring Bias in Vision Language Models on Gender, Race, Age, and Skin Tone

Modern neural translation models based on the Transformer architecture are known for their high performance, particularly when trained on high-resource datasets. A standard next-token prediction training strategy, while widely adopted in practice, may lead to overlooked artifacts such as representation collapse. Previous works have shown that this problem is especially pronounced in the representation of the deeper Transformer layers, where it often fails to efficiently utilize the geometric space. Representation collapse is even more evident in end-to-end training of continuous-output neural machine translation, where the trivial solution would be to set all vectors to the same value. In this work, we analyze the dynamics of representation collapse at different levels of discrete and continuous NMT transformers throughout training. We incorporate an existing regularization method based on angular dispersion and demonstrate empirically that it not only mitigates collapse but also improves translation quality. Furthermore, we show that quantized models exhibit similar collapse behavior and that the benefits of regularization are preserved even after quantization.

Representation Collapse in Machine Translation Through the Lens of Angular Dispersion

Multimodal models excel in English, supported by abundant imageвЂ“text and audioвЂ“text data, but performance drops sharply for other languages due to limited multilingual multimodal resources. Existing solutions rely on machine translation, while advances in multilingual text modeling remain underutilized. We introduce M2M, a lightweight alignment method that learns only a few linear layersвЂ”using English text aloneвЂ”to map multilingual text embeddings into multimodal space. Despite its simplicity, M2M matches baseline performance in English (94.9% Recall@10) and achieves strong zero-shot transfer (89.5% Recall@10 averaged across 11 languages, 10 unseen) on XTD Text-to-Image retrieval. Qualitative t-SNE visualizations show that multilingual embeddings align tightly with multimodal representations, while weight analysis reveals that the transformation reshapes embedding geometry rather than performing trivial rotations. Beyond imageвЂ“text retrieval, M2M generalizes to AudioвЂ“Text retrieval and cross-lingual Text-to-Image generation. We release code, checkpoints, and multilingual evaluation datasets (https://github.com/m2m-codebase/M2M) to facilitate further research.

Downloads

Next from EACL 2026

Plane Geometry Problem Solving with Multi-modal Reasoning: A Survey

Similar lecture

Improve Safety Training of Large Language Models with Safety-Critical Singular Vectors Localization

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES