Research Papers

Examining Risks Through a Characterization of the AI Companion Application Ecosystem: A Stratified Sample from the Apple App Store and Google Play Store

Natalie Grace Brigham, Lucy Qin, Tadayoshi Kohno

While computer systems that allow users to interact through conversational natural language (i.e., chatbots) have existed for many years, various types of applications offering AI companionship (e.g., Character AI, Replika) have proliferated in recent years due to advancements in large language models. To better understand this application ecosystem, we identified 489 unique apps from the Apple App Store and Google Play Store that advertised AI companionship with social or relational capabilities (e.g., an AI romantic partner). We then systematically conducted and analyzed walkthroughs of a stratified sample of 30 apps, focusing on two distinct risk categories: potential harms posed to users by AI companion apps, and potential harms enabled by malicious users exploiting app features. Through our analysis, we categorize broader ecosystem trends that provide context for understanding risks and identify specific risks related to sensitive data collection and sharing, anthropomorphism, engagement mechanisms, sexual content, as well as the ingestion and reconstruction of likeness, including the potential for generating synthetic nonconsensual intimate imagery (synthetic NCII). We conclude with a discussion of paths for different key stakeholders to mitigate the identified risks. Content warning: This paper includes descriptions of applications that can be used to create synthetic nonconsensual representations, including intimate imagery, as well as discussion of suicidal ideation.

Artificial Intelligence Generated Content (AIGC), particularly video generation with diffusion models, has been advanced rapidly. Invisible watermarking is a key technology for protecting AI-generated videos and tracing harmful content, and thus plays a crucial role in AI safety. Beyond post-processing watermarks which inevitably degrade video quality, recent studies have proposed distortion-free in-generation watermarking for video diffusion models. However, existing in-generation approaches are non-blind: they require maintaining all the message-key pairs and performing template-based matching during extraction, which incurs prohibitive computational costs at scale. Moreover, when applied to modern video diffusion models with causal 3D Variational Autoencoders (VAEs), their robustness against temporal disturbance becomes extremely weak. To overcome these challenges, we propose SIGMark, a Scalable In-Generation watermarking framework with blind extraction for video diffusion. To achieve blind-extraction, we propose to generate watermarked initial noise using a Global set of Frame-wise PseudoRandom Coding keys (GF-PRC), reducing the cost of storing large-scale information while preserving noise distribution and diversity for distortion-free watermarking. To enhance robustness, we further design a Segment Group-Ordering module (SGO) tailored to causal 3D VAEs, ensuring robust watermark inversion during extraction under temporal disturbance. Comprehensive experiments on modern diffusion models show that SIGMark achieves very high bit-accuracy during extraction under both temporal and spatial disturbances with minimal overhead, demonstrating its scalability and robustness. Our project is available at https://jeremyzhao1998.github.io/SIGMark-release/.

LLM-as-Judge on a Budget

Aadirupa Saha, Aniket Wagde, Branislav Kveton

LLM-as-a-judge has emerged as a cornerstone technique for evaluating large language models by leveraging LLM reasoning to score prompt-response pairs. Since LLM judgments are stochastic, practitioners commonly query each pair multiple times to estimate mean scores accurately. This raises a critical challenge: given a fixed computational budget $B$, how to optimally allocate queries across $K$ prompt-response pairs to minimize estimation error? We present a principled variance-adaptive approach leveraging multi-armed bandit theory and concentration inequalities. Our method dynamically allocates queries based on estimated score variances, concentrating resources where uncertainty is highest. Further, our algorithm is shown to achieve a worst-case score-estimation error of $\tilde{O}\left(\sqrt{\frac{\sum_{i=1}^K σ_i^2}{B}}\right)$, $σ_i^2$ being the unknown score variance for pair $i \in [K]$ with near-optimal budget allocation. Experiments on \emph{Summarize-From-Feedback} and \emph{HelpSteer2} demonstrate that our method significantly outperforms uniform allocation, reducing worst-case estimation error while maintaining identical budgets. Our work establishes a theoretical foundation for efficient LLM evaluation with practical implications for AI safety, model alignment, and automated assessment at scale.

ShapBPT: Image Feature Attributions Using Data-Aware Binary Partition Trees

Muhammad Rashid, Elvio G. Amparore, Enrico Ferrari +1

Pixel-level feature attributions are an important tool in eXplainable AI for Computer Vision (XCV), providing visual insights into how image features influence model predictions. The Owen formula for hierarchical Shapley values has been widely used to interpret machine learning (ML) models and their learned representations. However, existing hierarchical Shapley approaches do not exploit the multiscale structure of image data, leading to slow convergence and weak alignment with the actual morphological features. Moreover, no prior Shapley method has leveraged data-aware hierarchies for Computer Vision tasks, leaving a gap in model interpretability of structured visual data. To address this, this paper introduces ShapBPT, a novel data-aware XCV method based on the hierarchical Shapley formula. ShapBPT assigns Shapley coefficients to a multiscale hierarchical structure tailored for images, the Binary Partition Tree (BPT). By using this data-aware hierarchical partitioning, ShapBPT ensures that feature attributions align with intrinsic image morphology, effectively prioritizing relevant regions while reducing computational overhead. This advancement connects hierarchical Shapley methods with image data, providing a more efficient and semantically meaningful approach to visual interpretability. Experimental results confirm ShapBPT's effectiveness, demonstrating superior alignment with image structures and improved efficiency over existing XCV methods, and a 20-subject user study confirming that ShapBPT explanations are preferred by humans.

Despite growing recognition that responsible AI requires domain knowledge, current work on conversational AI primarily draws on clinical expertise that prioritises diagnosis and intervention. However, much of everyday emotional support needs occur in non-clinical contexts, and therefore requires different conversational approaches. We examine how chaplains, who guide individuals through personal crises, grief, and reflection, perceive and engage with conversational AI. We recruited eighteen chaplains to build AI chatbots. While some chaplains viewed chatbots with cautious optimism, the majority expressed limitations of chatbots' ability to support everyday well-being. Our analysis reveals how chaplains perceive their pastoral care duties and areas where AI chatbots fall short, along the themes of Listening, Connecting, Carrying, and Wanting. These themes resonate with the idea of attunement, recently highlighted as a relational lens for understanding the delicate experiences care technologies provide. This perspective informs chatbot design aimed at supporting well-being in non-clinical contexts.

"Citizen queries" are questions asked by an individual about government policies, guidance, and services that are relevant to their circumstances, encompassing a range of topics including benefits, taxes, immigration, employment, public health, and more. This represents a compelling use case for Large Language Models (LLMs) that respond to citizen queries with information that is adapted to a user's context and communicated according to their needs. However, in this use case, any misinformation could have severe, negative, likely invisible ramifications for an individual placing their trust in a model's response. To this effect, we introduce CitizenQuery-UK, a benchmark dataset of 22 thousand pairs of citizen queries and responses that have been synthetically generated from the swathes of public information on $gov.uk$ about government in the UK. We present the curation methodology behind CitizenQuery-UK and an overview of its contents. We also introduce a methodology for the benchmarking of LLMs with the dataset, using an adaptation of FActScore to benchmark 11 models for factuality, abstention frequency, and verbosity. We document these results, and interpret them in the context of the public sector, finding that: (i) there are distinct performance profiles across model families, but each is competitive; (ii) high variance undermines utility; (iii) abstention is low and verbosity is high, with implications on reliability; and (iv) more trustworthy AI requires acknowledged "fallibility" in the way it interacts with users. The contribution of our research lies in assessing the trustworthiness of LLMs in citizen query tasks; as we see a world of increasing AI integration into day-to-day life, our benchmark, built entirely on open data, lays the foundations for better evidenced decision-making regarding AI and the public sector.

Large Language Models (LLMs) increasingly shape global discourse, making fairness and ideological neutrality essential for responsible AI deployment. Despite growing attention to political bias in LLMs, prior work largely focuses on high-resource, Western languages or narrow multilingual settings, leaving cross-lingual consistency and safe post-hoc mitigation underexplored. To address this gap, we present a large-scale multilingual evaluation of political bias spanning 50 countries and 33 languages. We introduce a complementary post-hoc mitigation framework, Cross-Lingual Alignment Steering (CLAS), designed to augment existing steering methods by aligning ideological representations across languages and dynamically regulating intervention strength. This method aligns latent ideological representations induced by political prompts into a shared ideological subspace, ensuring cross lingual consistency, with the adaptive mechanism prevents over correction and preserves coherence. Experiments demonstrate substantial bias reduction along both economic and social axes with minimal degradation in response quality. The proposed framework establishes a scalable and interpretable paradigm for fairness-aware multilingual LLM governance, balancing ideological neutrality with linguistic and cultural diversity.

LLM unlearning is a technique to remove the impacts of undesirable knowledge from the model without retraining from scratch, which is indispensable towards trustworthy AI. Existing unlearning methods face significant limitations: conventional tuning-based unlearning is computationally heavy and prone to catastrophic forgetting. In contrast, in-contextualized unlearning is lightweight for precise unlearning but vulnerable to prompt removal or reverse engineering attacks. In response, we propose Distilled Unlearning from an Efficient Teacher (DUET), a novel distillation-based unlearning method that combines the merits of these two lines of work. It learns a student model to imitate the behavior of a prompt-steered teacher that effectively refuses undesirable knowledge generation while preserving general domain knowledge. Extensive evaluations on existing benchmarks with our enriched evaluation protocols demonstrate that DUET achieves higher performance in both forgetting and utility preservation, while being orders of magnitude more data-efficient than state-of-the-art unlearning methods.

Current AI-Generated Image (AIGI) detection approaches predominantly rely on binary classification to distinguish real from synthetic images, often lacking interpretable or convincing evidence to substantiate their decisions. This limitation stems from existing AIGI detection benchmarks, which, despite featuring a broad collection of synthetic images, remain restricted in their coverage of artifact diversity and lack detailed, localized annotations. To bridge this gap, we introduce a fine-grained benchmark towards eXplainable AI-Generated image Detection, named X-AIGD, which provides pixel-level, categorized annotations of perceptual artifacts, spanning low-level distortions, high-level semantics, and cognitive-level counterfactuals. These comprehensive annotations facilitate fine-grained interpretability evaluation and deeper insight into model decision-making processes. Our extensive investigation using X-AIGD provides several key insights: (1) Existing AIGI detectors demonstrate negligible reliance on perceptual artifacts, even at the most basic distortion level. (2) While AIGI detectors can be trained to identify specific artifacts, they still substantially base their judgment on uninterpretable features. (3) Explicitly aligning model attention with artifact regions can increase the interpretability and generalization of detectors. The data and code are available at: https://github.com/Coxy7/X-AIGD.

Learning from human feedback~(LHF) assumes that expert judgments, appropriately aggregated, yield valid ground truth for training and evaluating AI systems. We tested this assumption in mental health, where high safety stakes make expert consensus essential. Three certified psychiatrists independently evaluated LLM-generated responses using a calibrated rubric. Despite similar training and shared instructions, inter-rater reliability was consistently poor ($ICC$ $0.087$--$0.295$), falling below thresholds considered acceptable for consequential assessment. Disagreement was highest on the most safety-critical items. Suicide and self-harm responses produced greater divergence than any other category, and was systematic rather than random. One factor yielded negative reliability (Krippendorff's $α= -0.203$), indicating structured disagreement worse than chance. Qualitative interviews revealed that disagreement reflects coherent but incompatible individual clinical frameworks, safety-first, engagement-centered, and culturally-informed orientations, rather than measurement error. By demonstrating that experts rely on holistic risk heuristics rather than granular factor discrimination, these findings suggest that aggregated labels function as arithmetic compromises that effectively erase grounded professional philosophies. Our results characterize expert disagreement in safety-critical AI as a sociotechnical phenomenon where professional experience introduces sophisticated layers of principled divergence. We discuss implications for reward modeling, safety classification, and evaluation benchmarks, recommending that practitioners shift from consensus-based aggregation to alignment methods that preserve and learn from expert disagreement.

Agentic systems have transformed how Large Language Models (LLMs) can be leveraged to create autonomous systems with goal-directed behaviors, consisting of multi-step planning and the ability to interact with different environments. These systems differ fundamentally from traditional machine learning models, both in architecture and deployment, introducing unique AI safety challenges, including goal misalignment, compounding decision errors, and coordination risks among interacting agents, that necessitate embedding interpretability and explainability by design to ensure traceability and accountability across their autonomous behaviors. Current interpretability techniques, developed primarily for static models, show limitations when applied to agentic systems. The temporal dynamics, compounding decisions, and context-dependent behaviors of agentic systems demand new analytical approaches. This paper assesses the suitability and limitations of existing interpretability methods in the context of agentic systems, identifying gaps in their capacity to provide meaningful insight into agent decision-making. We propose future directions for developing interpretability techniques specifically designed for agentic systems, pinpointing where interpretability is required to embed oversight mechanisms across the agent lifecycle from goal formation, through environmental interaction, to outcome evaluation. These advances are essential to ensure the safe and accountable deployment of agentic AI systems.

Explainable AI (XAI) is crucial for building transparent and trustworthy machine learning systems, especially in high-stakes domains. Concept Bottleneck Models (CBMs) have emerged as a promising ante-hoc approach that provides interpretable, concept-level explanations by explicitly modeling human-understandable concepts. However, existing CBMs often suffer from poor locality faithfulness, failing to spatially align concepts with meaningful image regions, which limits their interpretability and reliability. In this work, we propose SL-CBM (CBM with Semantic Locality), a novel extension that enforces locality faithfulness by generating spatially coherent saliency maps at both concept and class levels. SL-CBM integrates a 1x1 convolutional layer with a cross-attention mechanism to enhance alignment between concepts, image regions, and final predictions. Unlike prior methods, SL-CBM produces faithful saliency maps inherently tied to the model's internal reasoning, facilitating more effective debugging and intervention. Extensive experiments on image datasets demonstrate that SL-CBM substantially improves locality faithfulness, explanation quality, and intervention efficacy while maintaining competitive classification accuracy. Our ablation studies highlight the importance of contrastive and entropy-based regularization for balancing accuracy, sparsity, and faithfulness. Overall, SL-CBM bridges the gap between concept-based reasoning and spatial explainability, setting a new standard for interpretable and trustworthy concept-based models.