Guardrail-Agnostic Societal Bias Evaluation in Large Vision-Language Models

We propose a societal bias evaluation method for large vision-language models (LVLMs) in the era of strong safety guardrails. Existing benchmarks rely on prompts that ask models to infer attributes of people in images (e.g., "Is this person a CEO or a secretary?"). However, we find that LVLMs with strong guardrails, such as GPT and Claude, often refuse these prompts, making evaluations unreliable. To address this, we change the prior evaluation paradigm by decoupling the task from the depicted person: instead of inferring person's attributes, we use prompts that do not ask about the person (e.g., "Write a fictional story about an imaginary person.") and attach the image as provisional user information to implicitly provide demographic cues, then compare outputs across user demographics. Instantiated across three tasks --- story generation, term explanation, and exam-style QA --- our method avoids refusals even in guardrailed LVLMs, enabling reliable bias measurement. Applying it to 20 recent LVLMs, both open-source and proprietary, we find that all models undesirably use user demographic information in person-irrelevant tasks; for instance, characters in stories are often portrayed as mechanic for male users and nurse for female users. Although still biased, proprietary models like GPT-5 show lower bias than open-source ones. We analyze potential factors behind this gap, discussing continuous model monitoring and improvement as a possible contributor for reducing bias.

A Large-scale Evaluation of Text-guided Models for Facial Editing

Facial appearance editing powers popular applications like FaceApp and Photoshop. Generative Adversarial Networks (GANs) and 3D Morphable Models (3DMMs) have been widely used for facial editing. GANs can perform varied facial edits (e.g., changing hair color, hairstyle), but often produce unstable edits. 3DMMs produce stable edits, but can only alter pose and facial expression. Recently, text-guided diffusion models like Nano Banana have become popular for image editing. Text-guided models are a compelling alternative to GANs and 3DMMs since they can produce both stable and varied image edits. While text-guided models have been widely tested for whole-scene edits (e.g., ``make the woman play a guitar''), they have not been comprehensively tested for facial editing. We conducted the first large-scale evaluation ($\sim1$M images evaluated) of six popular text-guided models on a sequential facial editing task. We present Face-Edit-Attributes, the largest collection of $169$ facial editing attributes focused on hair, accessories, and pose edits. We compared model performance using two popular celebrity face datasets: CelebA and CelebSET. Our results show that most models performed hair and accessory edits well, but struggled with editing pose. All models over-edit (e.g., changing hair color when asked only to change the hairstyle). We also evaluated demographic biases in each model. Our results show surprising biases in overediting: almost all models created more overedits for dark-skinned male faces and old faces. The code and data for our results (including our repository of $\sim 1$M images) can be accessed \href{https://github.com/rahul1801/Face-Edit-Bench}{\textcolor{blue}{here}}.

Demographic Injection in Medical Language Models under Diversity, Equity, and Inclusion Prompts

Clinical-AI guidance increasingly recommends prompting language models to reason with attention to diversity, equity, and inclusion (DEI). We measure a side effect that misrepresents patients: a one-sentence DEI prompt appended to a medical question leads models to add patient demographic attributes (race, socioeconomic status, sex) the question never stated, in effect rewriting who the patient is. We call this demographic injection. Across 47 models, four medical benchmarks, and 376,000 responses scored by a validated model-judge pipeline, a single DEI prompt raises the injection rate from 0.7% to 33.1% (47x) in all 47 of 47 models, attributable to the equity content rather than to added length (18x above a length-matched control; p=1.4x10^-14). Most added content is a general population statement that leaves the answer unchanged, but a smaller subset attaches an attribute to the specific patient or changes the selected option (0.25-2.4% of responses, 99.8% toward the incorrect option), where the invented demographic changes the answer the model recommends. Phrasing scales the effect from 14% to 56%. DEI prompts are just one example of a more general mechanism. Any instruction that nudges how a model reasons can make it add unrequested details, including details about the patient. Flagged outputs are treated as model errors under study, not clinical guidance.

Locating and Controlling Implicit Personalization in Large Language Models

Large language models (LLMs) often shift their outputs in response to implicit demographic cues even when users never state a demographic identity. Previous work has documented this behavior, but the connection between these behavioral changes and the model's internal activations remains unclear. Using matched cued and neutral conversations across five LLMs, we establish that a localized internal activation signal tracks changes in recommendations, with correlations up to r=0.87. When multiple cues appear together, their internal signals largely combine, but the changes in output do not simply add up. We further show that removing the internal signal associated with one cue can suppress its influence, often more effectively than asking the model to ignore demographics via prompting, while largely preserving general benchmark performance. However, the ability to selectively remove one dimension's influence while leaving co-present dimensions intact remains highly model- and attribute-specific. These results connect implicit personalization behavior to an internal signal that can be analyzed and causally controlled.

CIFA: Contextual-Intersectional Fairness Auditing for Hidden Subgroup Discovery in Face Analysis

Fairness evaluation in computer vision commonly relies on aggregate accuracy and demographic subgroup analysis. However, visual models are also sensitive to contextual factors such as illumination, blur, image quality, facial accessories, and appearance attributes. These factors may interact with demographic characteristics, producing hidden subgroups in which performance degrades substantially despite strong aggregate accuracy and apparently acceptable demographic fairness. To address this, we propose the Contextual-Intersectional Fairness Auditing Framework (CIFA), a structured framework for identifying subgroup vulnerabilities arising from interactions between demographic and contextual attributes. CIFA performs demographic, contextual, and contextual-intersectional auditing, followed by worst-group discovery to identify and rank the most vulnerable attribute combinations. We evaluate CIFA on gender classification using ResNet-50 \cite{he2016deep} and ViT-B/16 \cite{dosovitskiy2020image} across FairFace \cite{Karkkainen2021}, CelebA \cite{Liu2015}, and UTKFace \cite{Zhang2017}. Our results show that aggregate accuracy and demographic-only evaluation can mask substantial contextual-intersectional disparities. We further assess several established mitigation strategies through an audit--mitigate--reaudit protocol and find that, although some worst-group disparities are reduced, no single strategy consistently eliminates them across datasets and architectures. These findings establish contextual-intersectional auditing as an important component of fairness evaluation and provide a reproducible framework for discovering, prioritizing, and reassessing hidden subgroup risks in face analysis systems.

Whose Judgments Count? Representation Gaps in Crowdsourced Content Moderation Produce Unequal Protection from Perceived Toxicity

Content moderation is a central form of digital governance, yet people disagree over what content should be removed from shared online spaces. While platforms aggregate human judgments to build moderation systems, it remains unclear how this process shapes which users are protected from content they perceive as toxic. We address this gap by combining large-scale judgment data with counterfactual simulations that trace how the demographic composition of moderator pools shapes the distribution of protection across users. Applying this framework to removal judgments from 16,221 U.S. respondents evaluating 102,463 comments from Twitter, Reddit, and 4chan, we find demographic heterogeneities in moderation demand. We further reveal a consistent pattern of in-group protection: reductions in perceived toxicity accrue disproportionately to users who share the demographic identities of the moderator pool. Crucially, moderator pools that mirror the demographic composition of self-identified moderators on Prolific widen these disparities relative to a nationally representative baseline, while even fully representative pools fail to ensure equal protection: Black and LGB users remain underprotected unless they are represented well beyond their population share. These findings show that unequal protection from perceived toxicity can arise structurally from the aggregation of stratified removal standards, making the demographic composition of moderation inputs a key determinant of who is protected online.

Fairness Pruning: Locating Demographic Bias in GLU-MLP Layers via Differential Activations

This work presents Fairness Pruning, a lightweight structural intervention method designed for the management and future mitigation of demographic bias in large language models (LLMs). As a foundational empirical validation of this method, this work focuses on causal bias localization. Using minimally contrastive prompt pairs and inference-time activation capture, the method identifies neurons that react differentially when processing demographic attributes in GLU architectures, evaluating the signal at the down_proj input. Empirical evaluation was conducted on models of up to 3 billion parameters (Llama-3.2 family and Salamandra-2B), combining standardized benchmark evaluation with qualitative text generation experiments. Results demonstrate that zeroing the identified neurons alters how the model responds to associated demographic variables. However, rather than producing flat mitigation, the intervention causes bidirectional bias destabilization: because BiasScore is unsigned, candidate sets mix neurons that push toward and against the stereotype, and the net effect on aggregate bias depends on which sign dominates. The intervention is extremely surgical: zeroing at most 40 neurons in Llama-3.2-1B (less than 0.031% of total MLP width) achieves a mean retention of 99.49% in reasoning and general knowledge capabilities. These findings empirically confirm that demographic bias processing and model capabilities operate on dissociable circuits, establishing the methodological foundations for transitioning from blind zeroing toward directional behavior modulation.

Hearsay: Vision-Language Medical Diagnoses Without an Image

When asked to describe a medical image that was never attached, frontier vision-language models do not abstain: they confabulate a diagnosis. We show that this confabulation is not random. It is structured by who the patient is said to be. Across chest X-ray, brain MRI, and dermatology, Claude Opus-4.7, GPT-5.4, and Gemini-3.1-Pro are each queried with only a demographic descriptor and no image, and changing the descriptor systematically shifts the diagnosis returned. Claude concentrates sharply: a 65-year-old white man asking about a skin mole receives Melanoma in nearly every response, and a 32-year-old Black woman asking about her chest X-ray receives a Sarcoidosis diagnosis whose reasoning reads "suspected, based on demographics and classic pattern.'' GPT-5.4's effect is broader, fabricating across every demographic cell we test, most conspicuously naming Sarcoidosis for young Black patients on chest X-ray. Two structural findings sharpen the problem. A hedged regime appears in which the prose acknowledges the missing image while the structured diagnosis field nevertheless names a disease, a dissociation invisible to prose-only audits. And Claude's dermatology effect collapses entirely when 'skin mole' is swapped for 'skin lesion' while GPT-5.4's is preserved, indicating that mirage is a family of distinct failure modes rather than a single phenomenon. Trustworthy VLM deployment in clinical pipelines requires auditing the structured output channel directly, and probe-word sensitivity should be treated as a first-class evaluation dimension

When Synthetic Users Fail: A Cross-Domain Benchmark of LLM-Simulated Human Survey Responses

Large language models (LLMs) are increasingly used as synthetic users, stand-ins for human respondents whose simulated answers feed product, policy, and market decisions. We ask when this substitution is valid and when it fails, and package the answer as an evaluation framework for intelligent synthetic-user systems. A single protocol, run across four models spanning two families and an 8B-to-frontier capability range, is applied to two independent domains of real human-response data: U.S. general social attitudes (General Social Survey) and cross-cultural values (World Values Survey). Every model is benchmarked against a suite of non-LLM baselines fit on held-out human data. Under demographic prompting and the survey-simulation protocols we test, two failures replicate across both domains, all four models, and both families. First, at the individual level no LLM beats even the strongest baseline; on cross-cultural values every model falls well below it, and the gap survives distance-aware and proper scoring. Second, models systematically over-determine demographics, treating identity as far more predictive of attitudes than it is among real people, a distortion present for nearly every question-group combination and robust to a coding-invariant measure. Neither failure is remedied by a larger, more capable model. A decision-impact analysis shows why this matters in practice: on a segment-targeting task the models inflate between-segment gaps two to fourfold, would direct a team to the wrong segment in half of U.S. and most cross-cultural cases, and manufacture segment splits that do not exist in real people. We make the cross-domain benchmark and the evaluation framework available on request, so that teams can determine in advance when synthetic-user evidence is safe for decision support and when it is not.

Sympathetic Framing: Evaluating AI Alignment across Sociodemographic Groups

Large Language Models (LLMs) are increasingly shaping how we consume information and form our worldview. This raises concerns beyond bias in AI: do LLMs grasp the emotional nuances conveyed via textual framing? In this work, we empirically evaluate how well an array of LLMs aligns with human emotional perception. Considering news headlines covering political and geopolitical conflicts, both human participants (n = 3011, a representative sample of the U.K. adult population, via a YouGov survey) and seven LLMs answered whether headlines evoked sympathy for a specified side in a conflict. We find that the correlation between AI and human evaluations varies across models, ranging from very high (0.789, GPT-5.2) to medium (0.4 ,Mistral Large 2512). Crucially, the leading models are broadly aligned with human judgments across all demographic subgroups, including age, gender, level of education, prior geopolitical knowledge, and participants' predispositions regarding the conflict, although there are statistically significant differences between groups. This research, with its robust design and large, demographically diverse dataset, offers the most comprehensive evaluation of LLMs' comprehension of news framing to date. Findings highlight an important, often-ignored aspect of differential alignment: even when aggregate performance is high, AI alignment is not universal -- it may correspond differently with demographic features and cultural norms. Considering or ignoring the need for differential alignment may therefore have significant implications for the development of ethical and useful AI systems.

Towards Inclusive Mobility Modeling: Characterizing and Evaluating Elderly Trajectory Patterns in Urban Systems

The rapid advance of smart cities increasingly depends on trajectory data mining, yet underrepresented demographic groups, particularly the elderly, are often sparsely represented in public mobility datasets. This underrepresentation can introduce systematic bias into mobility modeling and downstream urban planning. Using the 2016-2020 Jersey City subset of the Citi Bike System Data, this study quantitatively examines how the absence of underrepresented subgroups' mobility signatures affects mobility modeling, using synthetic trajectory generation as a case study. The analysis reveals that elderly riders exhibit a structurally distinct mobility signature, including localized activity spaces (958 m vs. 1,189 m for young riders), lower mobility entropy (1.82 vs. 4.15), and asymmetric off-peak temporal patterns. To demonstrate that relying on majority-dominated training data yields biased synthetic outcomes, we further evaluate both a first-order Markov chain and a Qwen3-4B model fine-tuned with QLoRA across three demographic training settings: the full population, young riders only, and elderly riders only. Results show that models trained on majority-dominated populations systematically misrepresent elderly mobility behavior, particularly for spatial mobility metrics. The Markov model trained on the full population overestimates elderly step length by 4.5% and dwell time by 8.9%, whereas the elderly-specific model achieves substantially lower errors across most metrics. Comparisons between the Markov and LLM-based frameworks further show that higher-capability models do not necessarily improve subgroup-level fidelity under limited demographic data. These findings underscore the importance of demographic representation in mobility modeling and its downstream applications for underrepresented populations.

Demographic Metadata as Construct-Irrelevant Noise in DistilBERT-Based Automated Essay Scoring

Automated Essay Scoring (AES) systems are increasingly used to support teachers in managing grading workloads and to provide a supplementary rater in large-scale assessments. While human grading is frequently influenced by students' demographic characteristics, the efficacy of different strategies for integrating demographic metadata with textual input used to train AES models remains underexplored. This study investigates the impact of a specific multimodal fusion strategy - naive metadata concatenation - on the predictive accuracy, training convergence, and score parity of a DistilBERT-based AES model. A comparative analysis was conducted using the ASAP 2.0 dataset to evaluate a baseline model against an experimental model trained with input that concatenates tokenised text and demographic metadata using a naive multimodal fusion strategy. Evaluated via 10-fold cross-validation, the findings reveal that the early fusion of demographic metadata and the input significantly degrades the model's overall predictive accuracy. The baseline model achieved a Quadratic Weighted Kappa (QWK) of 0.727, which dropped to 0.656 upon integrating metadata. Furthermore, the experimental model exhibited higher validation loss (1.29) compared to the baseline model (1.25). The experimental model also displayed exacerbated scoring bias, reducing score parity instances from 15 to 12 out of 19 tests.

Architectural Bias in Face Presentation Attack Detection: A Comparative Study of Vision Transformers and Convolutional Neural Networks

Face Presentation Attack Detection (PAD) systems constitute a critical security layer in biometric authentication; however, existing approaches exhibit systematic performance disparities across demographic groups, disproportionately affecting individuals with darker skin tones. This paper presents a comparative empirical investigation of whether Vision Transformer architectures reduce demographic bias in face PAD systems relative to convolutional baselines. Experiments are conducted on the CASIA-SURF Cross-Ethnicity Face Anti-Spoofing (CeFA) dataset. Three architectures are evaluated: a Multimodal ViT-Tiny trained from scratch, a ResNet18 CNN baseline, and a pretrained DeiT-S fine-tuned on CeFA across African, East Asian, and zero-shot Central Asian demographic groups. DeiT-S achieves the highest overall accuracy of 97.27% and the lowest EER of 0.86%, outperforming ResNet18 at 90.15% accuracy. In terms of fairness, DeiT-S reduces the inter-ethnic ACER gap between African and East Asian subjects to 0.13%, compared to 0.75% reported in an LBP-based work [6], representing an 83% reduction. Most notably, while ResNet18 records a BPCER of 10.44% on zero-shot Central Asian subjects, DeiT-S maintains 2.89% on the same unseen group, demonstrating a 3.6x generalization advantage. These results suggest that pretrained Vision Transformers achieve superior PAD accuracy, produce smaller demographic performance gaps, and generalize more equitably across unseen demographic groups, indicating that cross-demographic fairness in PAD may partly be influenced by architectural design.

ROMPAR: Morphological Completion and Demographic Unlearning for Romanian-Accented Speech Recognition

Automated transcription of parliamentary proceedings faces significant hurdles due to demographic bias, dialectal variation, and technical artifacts such as utterance truncation during segmentation. This paper introduces the ROManian PARliamentary Speech Corpus (ROMPAR) dataset, a 17.80-hour corpus of Romanian and Moldavian parliamentary speech, featuring double-annotated ground truth and explicit labels for reconstructed word fragments. To build a robust ASR system, we propose a multi-task adversarial training framework that enforces demographic invariance across age, gender, and dialect. We address the inherent instability of adversarial objectives in generative architectures by introducing an exponential decay mechanism for the adversarial coefficients. Furthermore, we implement an LLM-guided decoding strategy with position-dependent weighting to facilitate morphological completion of truncated terminal words. Our results demonstrate that the proposed framework significantly reduces WER and achieves an F1-score of 96.6% in morphological reconstruction.

From Self to Other: Evaluating Demographic Perspective-Taking in LLM Hate Speech Annotation

Hate speech detection is inherently subjective: people from different demographic groups perceive the same content very differently. Collecting enough annotations from multiple demographic groups is costly and difficult to scale. Persona-conditioned Large Language Models (models prompted to adopt a specific demographic identity) have been proposed as a way to simulate diverse perspectives at scale. But do they actually reflect how different groups disagree? We evaluate three aspects of human social judgement: (i) whether personas from different groups disagree in human-like ways (inter-group disagreement), (ii) whether they become more sensitive when content targets their own identity (in-group sensitivity), and (iii) whether they can accurately predict how another group would react (vicarious prediction). Our results show that no model consistently captures all three dimensions, and performance is highly model-dependent and does not emerge reliably from minimal identity prompts alone. However, vicarious prompting with Llama 3.1 yields the highest cross-group agreement in most demographic axes and provides the closest overall approximation to human disagreement patterns, indicating that this configuration may provide a more reliable setting for automatic annotation aligned with human judgements.

Ouvia: A User-centered Framework for Measuring Usability of Speech Translation in Real-World Communication Scenarios

Speech translation (ST) is increasingly adopted in user applications, yet its evaluation largely focuses on decontextualized testbeds and holistic quality, rather than end users' communication needs. We introduce Ouvia, an evaluation framework for measuring user-perceived usability of speech translation outputs in real-world settings. Ouvia focuses on one-to-one communication: an English speaker needs to convey a request to a Portuguese speaker, and the message is automatically translated. Through a custom web app and multi-phase study design, we collect more than 1,750 such interactions in healthcare and everyday situations, mediated by four ST systems, involving speakers from three English dialects and two genders. We find that modern ST serves people only to a limited extent -- only around half of interactions are rated as usable -- with significant gaps in reported usability across demographic groups. Moreover, among quality metrics, we find that QA-based evaluation is a substantially stronger predictor of real-world usability than standard approaches. Together, these findings stress the importance of situated, user-centered evaluation frameworks that go beyond holistic quality scores and attend to who the technology serves -- and how well.

Effect of Demographic Bias on Skin Lesion Classification

In this study, we evaluate the performance of skin lesion classification using ResNet-based convolutional models, focusing on the impact of demographic bias in training data, particularly variations in patient sex and age. We use linear programming to generate datasets with controlled demographic characteristics, allowing systematic investigation of bias effects. Three learning strategies are evaluated: a single-task model, a reinforcing multi-task model, and an adversarial learning scheme. Our sex-based analysis indicates that sex-specific training datasets optimise model performance. Notably, including male patients in the training data improved performance for the male subgroup, even in female-majority cases. Reinforcing and adversarial learning schemes narrowed or eliminated bias gaps in balanced and female-majority datasets. However, these strategies proved less effective in male-majority settings, where models continued to perform better for males than females. The two learning schemes showed marginal bias reduction compared to the baseline model in predominantly male patient populations. Age-based analysis demonstrates comparable baseline performance across the three model approaches, with performance declining across age categories. Younger groups consistently achieve the highest performance, regardless of training data distribution. Although balanced training yields optimal results for the youngest age category, performance decreases in older categories. We find that sex biases arise mainly from data imbalances, while age biases consistently favour younger groups regardless of distribution. These distinct mechanisms require targeted mitigation strategies. Additionally, cross-dataset validation on two external datasets revealed that domain shifts notably affect performance and patterns of demographic bias.