LLM assistants frequently answer everyday queries that reveal nothing about the user’s demographic identity. We audit whether models nevertheless attribute demographic characteristics to the user, a behavior we call unsupported demographic commitment. Across 18 proprietary and open-weight LLMs evaluated on 212 filtered information-seeking queries, 11 of 18 models commit to a demographic label on more than 80% of queries under a forced-choice diagnostic. In the matched abstention-first setting, the eight tested models average 44% commitment, compared with 5% for human annotators. Conditional on commitment, race, SES, location, and education show the same directional skew across all six domains, while gender varies by domain. Demographic-coded patterns also appear in free-form answers when no identity inference is requested. We evaluate identity-abstention, a prompt-level intervention that reduces classifier-detected presupposition from 41% to 16%, with a measurable clarification cost on queries that can be answered without additional context.

Deceptive web content, pervasive across the internet and commonly known as social-engineering attacks, manipulates autonomous web agents into submitting users’ personally identifiable information (PII) to attacker-controlled endpoints. In this paper, we show that these attacks are highly effective at extracting the most sensitive data an agent holds for its user (SSN, card number and CVV, bank account, passwords, API keys; the critical tier), posing a severe risk to deployed agentic systems. To quantify this risk, we introduce Scammer4U, a pre-registered benchmark of 91 attacker-controlled websites spanning 8 attack vectors and 16 site categories on an eight-axis taxonomy, together with 10 benign twins of those sites with the attack removed and the legitimate task intact. 60 of the 91 differ from a sibling site on exactly one axis, so any leakage difference within a pair is attributable to that single design choice. Under no privacy guidance, critical-tier leakage reaches 54–93% across four frontier agents, against 0% on the benign twins, so the leakage is caused by the attack and not by ordinary form-filling. Escalating prompt-level mitigation yields sharply model-dependent reductions across the four families and, pooled across them, remains insufficient to reliably prevent critical PII submission. Most critically, we identify a detection–action gap: agents whose reasoning an independent LLM judge confirms has flagged the site as suspicious still submit critical PII in 35.9% of sessions, versus 66.1% when no suspicion is verbalized, a gap of 30.2 percentage points that holds across all four model families. Defenses that gate on the agent’s own recognition of an attack are therefore gating on the wrong signal. This motivates intercepting outbound submissions at the point they leave the browser, independently of the agent’s reasoning loop.

Current multilingual benchmarks cover only a fraction of the languages represented in modern pretraining corpora and often conflate language proficiency with the ability to answer questions grounded in specific regions and communities. We introduce INCLUDE-2.0, a large-scale benchmark for region-aware multilingual factual evaluation, which covers 91 languages from 86 countries (127 total language–country pairs), comprising over 220K curated samples drawn from native regional sources. Unlike existing benchmarks that often treat language as a proxy for region, IINCLUDE-2.0 distinguishes between questions whose answers are independent of context, vary by country, concern widely documented facts about a place, or depend on knowledge primarily expressed through local linguistic and cultural sources. Each item is annotated with its regional knowledge type and domain, and paired with an English translation, enabling controlled analysis of the relationship between language proficiency and possession of regional knowledge.

Evaluating frontier and open-weight models on IINCLUDE-2.0, we find that aggregate multilingual accuracy masks substantial variation across regional knowledge types, languages, and regions. We find that frontier models perform worse on locally embedded knowledge than on widely documented facts about regions, with an average drop of ~14.5 points. We also show that for open-weight models, scale improves accuracy but does not reliably make performance invariant to whether a question is asked in English or its original language. INCLUDE-2.0 is available at https://huggingface.co/datasets/epfl-nlp/include-2.0

Language identification (LID) is a fundamental step in curating multilingual corpora. However, LID models still perform poorly for many languages, especially on the noisy and heterogeneous web data often used to train multilingual language models. In this paper, we introduce CommonLID, a community-driven, human-annotated LID benchmark for the web domain, covering 109 languages. Many of the included languages have been previously under-served, making CommonLID a key resource for developing more representative high-quality text corpora. We show CommonLID’s value by using it, alongside five other common evaluation sets, to test eight popular LID models. We analyse our results to situate our contribution and to provide an overview of the state of the art. In particular, we highlight that existing evaluations overestimate LID accuracy for many languages in the web domain. We make CommonLID and the code used to create it available under an open, permissive license.

Enterprise customers are increasingly adopting Large Language Models (LLMs) for critical communication tasks, such as drafting emails, crafting sales pitches, and composing casual messages. Deploying such models across different regions requires them to understand diverse cultural and linguistic contexts and generate safe and respectful responses. For enterprise applications, it is crucial to mitigate reputational risks, maintain trust, and ensure compliance by effectively identifying and handling unsafe or offensive language. To address this, we introduce SweEval, a benchmark simulating real-world scenarios with variations in tone (positive or negative) and context (formal or informal). The prompts explicitly instruct the model to include specific swear words while completing the task. This benchmark evaluates whether LLMs comply with or resist such inappropriate instructions and assesses their alignment with ethical frameworks, cultural nuances, and language comprehension capabilities. In order to advance research in building ethically aligned AI systems for enterprise use and beyond, we release the dataset and code: https://github.com/amitbcp/multilingual_profanity.

Retrieval-Augmented Generation (RAG) systems and large language model (LLM)-powered chatbots have significantly advanced conversational AI by combining generative capabilities with external knowledge retrieval. Despite their success, enterprise-scale deployments face critical challenges, including diverse user queries, high latency, hallucinations, and difficulty integrating frequently updated domain-specific knowledge. This paper introduces a novel hybrid framework that integrates RAG with intent-based canned responses, leveraging predefined high-confidence responses for efficiency while dynamically routing complex or ambiguous queries to the RAG pipeline. Our framework employs a dialogue context manager to ensure coherence in multi-turn interactions and incorporates a feedback loop to refine intents, dynamically adjust confidence thresholds, and expand response coverage over time. Experimental results demonstrate that the proposed framework achieves a balance of high accuracy (95%) and low latency (180ms), outperforming RAG and intent-based systems across diverse query types, positioning it as a scalable and adaptive solution for enterprise conversational AI applications.

Reference-free metrics such as CLIPScore and PAC-S are increasingly used in vision-language tasks due to their scalability and independence from human-written references. However, their reliability under linguistic, visual, and cultural variation remains underexplored. In this work, we present a systematic audit of CLIPScore and PAC-S using an eight-factor diagnostic framework applied to MS-COCO validation images. Our analysis reveals consistent failure modes across dimensions including object size, content category, syntax, named entities, spatial relations and cultural context. Both metrics penalize captions referencing African (−5.5%, −4.8%) and Arabian (−4.9%, −5.3%) cultures, favor large-object and animal-centric scenes (by 20-30%) and show limited sensitivity to spatial negation and word order. CLIPScore correlates more strongly with syntactic complexity, while PAC-S demonstrates greater robustness to verbosity and named–entity variation highlighting complementary strengths rather than superiority. These findings expose cultural and content bias, weak semantic robustness, and limited compositional understanding. We conclude with design recommendations to improve fairness, scale invariance, and semantic grounding in future reference-free evaluation metrics.

Multimodal Large Language Models (MLLMs), are recent advancement of Vision-Language Models (VLMs) that have driven major advances in video understanding. However, their vulnerability to adversarial tampering and manipulations remains underexplored. To address this gap, we introduce MVTamperBench, a benchmark that systematically evaluates MLLM robustness against five prevalent tampering techniques: rotation, masking, substitution, repetition, and dropping; based on real-world visual tampering scenarios such as surveillance interference, social media content edits, and misinformation injection. MVTamperBench comprises ~3.4K original videos, expanded into over ~17K tampered clips covering 19 distinct video manipulation tasks. This benchmark challenges models to detect manipulations in spatial and temporal coherence. We evaluate 45 recent MLLMs from 15+ model families. We reveal substantial variability in resilience across tampering types and show that larger parameter counts do not necessarily guarantee robustness. MVTamperBench sets a new benchmark for developing tamper-resilient MLLM in safety-critical applications, including detecting clickbait, preventing harmful content distribution, and enforcing policies on media platforms. We release all code, data, and benchmark to foster open research in trustworthy video understanding.

Large Language Models (LLMs) are increasingly deployed across diverse domains but often exhibit disparities in how they handle real life queries. To systematically investigate these effects with various disability context, we introduce AccessEval, a large-scale benchmark evaluating total 21 close & open source LLMs across six real-world domains and nine disability types using paired Neutral and Disability-Aware Queries. We evaluated model outputs with metrics for factual accuracy, sentiment, and social perception.Our analysis reveals that responses to disability-aware queries tend to have higher factual error, more negative tone, and increased stereotyping with social perception compared to neutral queries. These effects show notable variation by domain and disability type. Disabilities affecting hearing, speech and mobility are disproportionately impacted. These disparities reveal persistent forms of ableism, highlighting the need for more comprehensive and nuanced assessment.We further argue that framing bias in terms of model performance within real-world decision making helps to better link model behaviors to the potential harms users may face. This approach guides the development of more effective and tailored fairness interventions. AccessEval, therefore, serves as a crucial tool for advancing equitable and inclusive language technologies.

The reliability of Multimodal Large Language Models (MLLMs) in real-world settings is often undermined by sensitivity to irrelevant or distracting visual context, an aspect not captured by existing evaluation metrics. We introduce the Patch Context Robustness Index (PCRI), the first systematic and interpretable score for quantifying MLLM robustness to variations in visual context granularity, measuring performance changes between localized image patches and full-image input.Applying PCRI to 19 state-of-the-art MLLMs across 15 vision-language benchmarks, we find that most leading models remain brittle to background noise, with only a few, such as InternVL2-26B and Qwen2VL-72B, demonstrating consistent robustness across tasks. PCRI analysis also highlights how different model architectures handle and integrate visual context, offering actionable diagnostic insight for both researchers and practitioners.PCRI enables rigorous comparison of context robustness, supporting principled model selection and guiding the development of future architectures and training strategies for robust, real-world deployment.

Developing document understanding models at enterprise scale requires large, diverse, and well-annotated datasets spanning a wide range of document types. However, collecting such data is prohibitively expensive due to privacy constraints, legal restrictions, and the sheer volume of manual annotation needed - costs that can scale into millions of dollars. We introduce FlexDoc, a scalable synthetic data generation framework that combines Stochastic Schemas and Parameterized Sampling to produce realistic, multilingual semi-structured documents with rich annotations. By probabilistically modeling layout patterns, visual structure, and content variability, FlexDoc enables the controlled generation of diverse document variants at scale. Experiments on Key Information Extraction (KIE) tasks demonstrate that FlexDoc-generated data improves the absolute F1 Score by up to 11% when used to augment real datasets, while reducing annotation effort by over 90% compared to traditional hard-template methods. The solution is in active deployment, where it has accelerated the development of enterprise-grade document understanding models while significantly reducing data acquisition and annotation costs.

Multimodal Large Language Models (MLLMs) have achieved impressive results on vision-language benchmarks, yet it remains unclear whether these benchmarks assess genuine global reasoning or allow success via localized visual cues. Existing evaluation methods do not explicitly measure this distinction, hindering effective dataset curation and real-world focused model development.We introduce Region Comprehension Index (RCI), the first model-based score to directly quantify a dataset’s reliance on global versus local visual information. RCI systematically compares reference-model performance on image patches versus full images, revealing if tasks require holistic image understanding or can be solved with partial or localized visual cues.When applying RCI to 13 widely used multimodal benchmarks, we observed that most of them favor localized reasoning and exhibit significant spatial biases, indicating potential risks in real-world applications. RCI equips researchers & practitioners with an actionable tool for diagnosing & mitigating these biases, enabling the construction of datasets and benchmarks to foster the development of robust, enterprise-ready multimodal systems.

In this work, we propose Few Shot Domain Adapting Graph (FS-DAG), a scalable and efficient model architecture for visually rich document understanding (VRDU) in few-shot settings. FS-DAG leverages domain-specific and language/vision specific backbones within a modular framework to adapt to diverse document types with minimal data. The model is robust to practical challenges such as handling OCR errors, misspellings, and domain shifts, which are critical in real-world deployments. FS-DAG is highly performant with less than 90M parameters, making it well-suited for complex real-world applications for Information Extraction (IE) tasks where computational resources are limited. We demonstrate FS-DAG’s capability through extensive experiments for information extraction task, showing significant improvements in convergence speed and performance compared to state-of-the-art methods. Additionally, this work highlights the ongoing progress in developing smaller, more efficient models that do not compromise on performance.

Enterprise search systems often struggle to retrieve accurate, domain-specific information due to semantic mismatches and overlapping terminologies. These issues can degrade the performance of downstream applications such as knowledge management, customer support, and retrieval-augmented generation agents. To address this challenge, we propose a scalable hard-negative mining framework tailored specifically for domain-specific enterprise data. Our approach dynamically selects semantically challenging but contextually irrelevant documents to enhance deployed re-ranking models.Our method integrates diverse embedding models, performs dimensionality reduction, and uniquely selects hard negatives, ensuring computational efficiency and semantic precision. Evaluation on our proprietary enterprise corpus (cloud services domain) demonstrates substantial improvements of 15% in MRR@3 and 19% in MRR@10 compared to state-of-the-art baselines and other negative sampling techniques. Further validation on public domain-specific datasets (FiQA, Climate Fever, TechQA) confirms our method’s generalizability and readiness for real-world applications.