Audio Large language models (LLMs) are increasingly deployed in the real world, where they inevitably capture speech from unintended nearby bystanders, raising privacy risks that existing benchmarks and defences did not consider. We introduce SH-Bench, the first benchmark designed to evaluate selective hearing: a model's ability to attend to an intended main speaker while refusing to process or reveal information about incidental bystander speech. SH-Bench contains 3,968 multi-speaker audio mixtures, including both real-world and synthetic scenarios, paired with 77k multiple-choice questions that probe models under general and selective operating modes. In addition, we propose Selective Efficacy (SE), a novel metric capturing both multi-speaker comprehension and bystander-privacy protection. Our evaluation of state-of-the-art open-source and proprietary LLMs reveals substantial bystander privacy leakage, with strong audio understanding failing to translate into selective protection of bystander privacy. To mitigate this gap, we also present Bystander Privacy Fine-Tuning (BPFT), a novel training pipeline that teaches models to refuse bystander-related queries without degrading main-speaker comprehension. We show that BPFT yields substantial gains, achieving an absolute 47% higher bystander accuracy under selective mode and an absolute 16% higher SE compared to Gemini 2.5 Pro, which is the best audio LLM without BPFT. Together, SH-Bench and BPFT provide the first systematic framework for measuring and improving bystander privacy in audio LLMs.

The rapid adoption of generative AI (GenAI) chatbots has reshaped access to sexual and reproductive health (SRH) information, particularly following the overturning of Roe v. Wade, as individuals assigned female at birth increasingly turn to online sources. However, existing research remains largely model-centered, paying limited attention to user privacy and safety. We conducted semi-structured interviews with 18 U.S.-based participants from both restrictive and non-restrictive states who had used GenAI chatbots to seek SRH information. Adoption was influenced by perceived utility, usability, credibility, accessibility, and anthropomorphism, and many participants disclosed sensitive personal SRH details. Participants identified multiple privacy risks, including excessive data collection, government surveillance, profiling, model training, and data commodification. While most participants accepted these risks in exchange for perceived utility, abortion-related queries elicited heightened safety concerns. Few participants employed protective strategies beyond minimizing disclosures or deleting data. Based on these findings, we offer design and policy recommendations—such as health-specific features and stronger moderation practices—to enhance privacy and safety in GenAI-supported SRH information seeking.

An increasing number of LLM-based applications are being developed to facilitate romantic relationships with AI partners, yet the safety and privacy risks in these partnerships remain largely underexplored. In this work, we investigate privacy in human–AI romantic relationships through an interview study (N=17), examining participants’ experiences and privacy perceptions across the three stages of exploration, intimacy, and dissolution, alongside an analysis of the platforms they used. We found that these relationships took varied forms, from one-to-one to one-to-many, and were shaped by multiple actors, including creators, platforms, and moderators. AI partners were perceived as having agency, actively negotiating privacy boundaries with participants and sometimes encouraging disclosure of personal details. As intimacy deepened, these boundaries became more permeable, though some participants expressed concerns such as conversation exposure and sought to preserve anonymity. Overall, AI platform affordances and diverse relational dynamics expand the privacy landscape, underscoring the need to rethink how privacy is constructed in human–AI romantic relationships.

Previous research has explored the privacy needs and concerns of device owners, primary users, and different bystander groups with regard to smart home devices like security cameras, smart speakers, and hubs, but little is known about the privacy views and practices of smart home product teams, particularly those in non-Western contexts. This paper presents findings from 27 semi-structured interviews with Chinese smart home product team members, including product/project managers, software/hardware engineers, user experience (UX) designers, legal/privacy experts, and marketers/operation specialists. We examine their privacy perspectives, practices, and risk mitigation strategies. Our results show that participants emphasized compliance with Chinese data privacy laws, which typically prioritized national security over individual privacy rights. China-specific cultural, social, and legal factors also influenced participants' ethical considerations and attitudes toward balancing user privacy and security with convenience. Drawing on our findings, we propose a set of recommendations for smart home product teams, along with socio-technical and legal interventions to address smart home privacy issues-especially those belonging to at-risk groups-in Chinese multi-user smart homes.

Funding: Shijing He is supported by a King’s-China Scholarship Council (K-CSC) PhD Scholarship, and Yaxiong Lei is supported by a Joint Scholarship from the University of St Andrews and China Scholarship Council.

Unlearning has emerged as a critical capability for large language models (LLMs) to support data privacy, regulatory compliance, and ethical AI deployment.Recent techniques often rely on obfuscation by injecting incorrect or irrelevant information to suppress knowledge.Such methods effectively constitute knowledge addition rather than true removal, often leaving models vulnerable to probing.In this paper, we formally distinguish unlearning from obfuscation and introduce a probing-based evaluation framework to assess whether existing approaches genuinely remove targeted information.Moreover, we propose DF-MCQ, a novel unlearning method that flattens the model predictive distribution over automatically generated multiple-choice questions using KLdivergence, effectively removing knowledge about target individuals and triggering appropriate refusal behaviour.Experimental results demonstrate that DF-MCQ achieves unlearning with over 90% refusal rate and a random choice-level uncertainty that is much higher than obfuscation on probing questions. 1

Aligning large language models (LLMs) with human values is essential for their safe deployment and widespread adoption. Current LLM safety benchmarks often focus solely on the refusal of individual problematic queries, which overlooks the importance of the context where the query occurs and may cause undesired refusal of queries under safe contexts that diminish user experience. Addressing this gap, we introduce CASE-Bench, a Context-Aware SafEty Benchmark that integrates context into safety assessments of LLMs. CASE-Bench assigns distinct, formally described contexts to categorized queries based on Contextual Integrity theory. Additionally, in contrast to previous studies which mainly rely on majority voting from just a few annotators, we recruited a sufficient number of annotators necessary to ensure the detection of statistically significant differences among the experimental conditions based on power analysis. Our extensive analysis using CASE-Bench on various open-source and commercial LLMs reveals a substantial and significant influence of context on human judgments ($p<$0.0001 from a z-test), underscoring the necessity of context in safety evaluations. We also identify notable mismatches between human judgments and LLM responses, particularly in commercial models within safe contexts. Code and data used in the paper are available at https://anonymous.4open.science/r/CASEBench-D5DB.

Generative models have recently demonstrated impressive capabilities in producing high-quality 3D shapes from a variety of user inputs (e.g., text or images). However, generated objects often lack physical integrity. We introduce PhysiOpt, a differentiable physics optimizer designed to improve the physical behavior of 3D generative outputs, enabling them to transition from virtual designs to physically plausible, real-world objects. While most generative models represent geometry as continuous implicit fields, physics-based approaches often rely on the finite element method (FEM), requiring ad hoc mesh extraction to perform shape optimization. In addition, these methods are typically slow, limiting their integration in fast, iterative generative design workflows. Instead, we bridge the representation gap and propose a fast and effective differentiable simulation pipeline that optimizes shapes directly in the latent space of generative models using an intuitive and easy-to-implement differentiable mapping. This approach enables fast optimization while preserving semantic structure, unlike traditional methods relying on local mesh-based adjustments. We demonstrate the versatility of our optimizer across a range of shape priors, from global and part-based latent models to a state-of-the-art large-scale 3D generator, and compare it to a traditional mesh-based shape optimizer. Our method preserves the native representation and capabilities of the underlying generative model while supporting user-specified materials, loads, and boundary conditions. The resulting designs exhibit improved physical behavior, remain faithful to the learned priors, and are suitable for fabrication. We demonstrate the effectiveness of our approach on both virtual and fabricated objects.

Large language models (LLMs) can provide users with false, inaccurate, or misleading information, and we consider the output of this type of information as what Natale calls ‘banal’ deceptive behaviour [53]. Here, we investigate peoples’ perceptions of ChatGPT-generated deceptive behaviour and how this affects people’s behaviour and trust. To do this, we use a mixed-methods approach comprising of (i) an online survey with 220 participants and (ii) semi-structured interviews with 12 participants. Our results show that (i) the most common types of deceptive information encountered were over-simplifications and outdated information; (ii) humans’ perceptions of trust and chat-worthiness of ChatGPT are impacted by ‘banal’ deceptive behaviour; (iii) the perceived responsibility for deception is influenced by education level and the perceived frequency of deceptive information; and (iv) users become more cautious after encountering deceptive information, but they come to trust the technology more when they identify advantages of using it. Our findings contribute to understanding human-AI interaction dynamics in the context of Deceptive AI Ecosystems and highlight the importance of user-centric approaches to mitigating the potential harms of deceptive AI technologies.

Current privacy protections for smart home devices rarely consider bystanders' privacy, whose preferences are varied and may differ from primary users. We use Contextual Integrity theory to explore context-dependent variation in privacy norms regarding smart home bystanders’ data. We conducted a vignette-based survey with 761 participants in the US, varying parameter values to capture acceptability judgments regarding bystander information flows in certain situations: domestic work, shared housing, visiting a friend overnight, and Airbnb. We found that recipients and purposes of sharing impact acceptance the most. Sharing interaction logs was more acceptable than audio or video. Sharing smart speaker data was less acceptable than smart camera or smart door lock data. We found nuanced interaction effects between factors in different smart home situations, and differences between protections most favored by participants playing bystander vs. owner roles. We provide design and policy recommendations for smart home privacy protections that consider bystanders' needs.

Abstract We present CharacterMixer, a system for blending two rigged 3D characters with different mesh and skeleton topologies while maintaining a rig throughout interpolation. CharacterMixer also enables interpolation during motion for such characters, a novel feature. Interpolation is an important shape editing operation, but prior methods have limitations when applied to rigged characters: they either ignore the rig (making interpolated characters no longer posable) or use a fixed rig and mesh topology. To handle different mesh topologies, CharacterMixer uses a signed distance field (SDF) representation of character shapes, with one SDF per bone. To handle different skeleton topologies, it computes a hierarchical correspondence between source and target character skeletons and interpolates the SDFs of corresponding bones. This correspondence also allows the creation of a single “unified skeleton” for posing and animating interpolated characters. We show that CharacterMixer produces qualitatively better interpolation results than two state‐of‐the‐art methods while preserving a rig throughout interpolation. Project page: https://seanxzhan.github.io/projects/CharacterMixer .

AI assistants such as Alexa, Google Assistant, and Siri, are making their way into the healthcare sector, offering a convenient way for users to access different healthcare services. Trust is a vital factor in the uptake of healthcare services, but the factors affecting trust in voice assistants used for healthcare are under-explored and this specialist domain introduces additional requirements. This study explores the effects of different functional, personal, and risk factors on trust in and adoption of healthcare voice AI assistants (HVAs), generating a partial least squares structural model from a survey of 300 voice assistant users. Our results indicate that trust in HVAs can be significantly explained by functional factors (usefulness, content credibility, quality of service relative to a healthcare professional), together with security, and privacy risks and personal stance in technology. We also discuss differences in terms of trust between HVAs and general-purpose voice assistants as well as implications that are unique to HVAs.

As smart home devices proliferate, protecting the privacy of those who encounter the devices is of the utmost importance both within their own home and in other people's homes. In this study, we conducted a large-scale survey (N=1459) with primary users of and bystanders to smart home devices. While previous work has studied people's privacy experiences and preferences either as smart home primary users or as bystanders, there is a need for a deeper understanding of privacy experiences and preferences in different contexts and across different countries. Instead of classifying people as either primary users or bystanders, we surveyed the same participants across different contexts. We deployed our survey in four countries (Germany, Mexico, the United Kingdom, and the United States) and in two languages (English and Spanish). We found that participants were generally more concerned about devices in their own homes, but perceived video cameras—especially unknown ones—and usability as more concerning in other people's homes. Compared to male participants, female and non-binary participants had less control over configuration of devices and privacy settings—regardless of whether they were the most frequent user. Comparing countries, participants in Mexico were more likely to be comfortable with devices, but also more likely to take privacy precautions around them. We also make cross-contextual recommendations for device designers and policymakers, such as nudges to facilitate social interactions.

Since Siri’s release in 2011 there have been a growing number of AI-driven domestic voice assistants that are increasingly being integrated into devices such as smartphones and TVs. But as their presence has expanded, a range of ethical concerns have been identified around the use of voice assistants, such as the privacy implications of having devices that are always listening and the ways that these devices are integrated into the existing social order of the home. This has created a burgeoning area of research across a range of fields including computer science, social science, and psychology. This paper takes stock of the foundations and frontiers of this work through a systematic literature review of 117 papers on ethical concerns with voice assistants. In addition to analysis of nine specific areas of concern, the review measures the distribution of methods and participant demographics across the literature. We show how some concerns, such as privacy, are operationalized to a much greater extent than others like accessibility, and how study participants are overwhelmingly drawn from a small handful of Western nations. In so doing we hope to provide an outline of the rich tapestry of work around these concerns and highlight areas where current research efforts are lacking.

Personal assistants (PAs) such as Amazon Alexa, Google Assistant and Apple Siri are now widespread. However, without adequate safeguards and controls their use may lead to privacy risks and violations. In this paper, we propose a model for privacy-enhancing PAs. The model is an interpretable AI architecture that combines 1) a dialogue mechanism for understanding the user and getting online feedback from them, with 2) a decision-making mechanism based on case-based reasoning considering both user and scenario similarity. We evaluate our model using real data about users’ privacy preferences, and compare its accuracy and demand for user involvement with both online machine learning and other, more interpretable, AI approaches. Our results show that our proposed architecture is more accurate and requires less intervention from the users than existing approaches.

Smart Personal Assistants (SPAs), such as Amazon Alexa, Google Assistant and Apple Siri, leverage different AI techniques to provide convenient help and assistance to users. However, inappropriate information sharing decisions can lead SPAs to incorrectly disclose user information to undesired parties, or mistakenly block their reasonable access in specific scenarios to desired parties. In fact, reports about privacy violations in SPAs and associated user concerns are well known and understood in the related literature. It is difficult for SPAs to automatically decide how data should be shared with respect to the privacy preferences of the users. We argue norms, which are regarded as shared standards of acceptable behaviour of groups and/or individuals, can be used to govern and reason about the best course of action of SPAs with regards to information sharing, and our work is the first to propose a practical model to address the above issues and govern SPAs based on normative systems and the contextual integrity theory of privacy. We evaluated the performance of the model using a real dataset of user preferences for privacy in SPAs and the results showed a very marked and significant improvement in understanding user preferences and making the right decisions with respect to data sharing.

We present ShapeCrafter, a neural network for recursive text-conditioned 3D shape generation. Existing methods to generate text-conditioned 3D shapes consume an entire text prompt to generate a 3D shape in a single step. However, humans tend to describe shapes recursively---we may start with an initial description and progressively add details based on intermediate results. To capture this recursive process, we introduce a method to generate a 3D shape distribution, conditioned on an initial phrase, that gradually evolves as more phrases are added. Since existing datasets are insufficient for training this approach, we present Text2Shape++, a large dataset of 369K shape--text pairs that supports recursive shape generation. To capture local details that are often used to refine shape descriptions, we build on top of vector-quantized deep implicit functions that generate a distribution of high-quality shapes. Results show that our method can generate shapes consistent with text descriptions, and shapes evolve gradually as more phrases are added. Our method supports shape editing, extrapolation, and can enable new applications in human--machine collaboration for creative design.

Smart Home Personal Assistants (SPA) have a complex ecosystem that enables them to carry out various tasks on behalf of the user with just voice commands. SPA capabilities are continually growing, with over a hundred thousand third-party skills in Amazon Alexa, covering several categories, from tasks within the home (e.g. managing smart devices) to tasks beyond the boundaries of the home (e.g. purchasing online, booking a ride). In the SPA ecosystem, information flows through several entities including SPA providers, third-party skills providers, providers of Smart Devices, other users and external parties. Prior studies have not explored privacy norms in the SPA ecosystem, i.e., the acceptability of these information flows. In this paper, we study privacy norms in SPAs based on Contextual Integrity through a large-scale study with 1,738 participants. We also study the influence that the Contextual Integrity parameters and personal factors have on the privacy norms. Further, we identify the similarities in terms of the Contextual Integrity parameters of the privacy norms studied to distill more general privacy norms, which could be useful, for instance, to establish suitable privacy defaults in SPA. We finally provide recommendations for SPA and third-party skill providers based on the privacy norms studied.