Papers

Clinician review of every AI output is often proposed as a safeguard in mental healthcare, but vigilance research suggests this approach fails at scale and may paradoxically reduce safety. Drawing on our experience deploying an AI coaching tool across 350,000+ conversations between therapy sessions, we describe how we arrived at a three-layer human-on-the-loop oversight framework combining preventive design, real-time monitoring, and continuous clinician evaluation. We show how specific findings from clinical review drove iterative improvements, and offer practical recommendations for mental health professionals evaluating AI systems.

Algorithmic outputs now populate the digital environments through which contemporary life is organized. The role of law in facilitating and constituting (rather than merely responding to) these processes is gaining increasing traction across scholarly accounts. This inquiry traces the evolution of algorithmic outputs attending to their legal underpinnings and social implications, surfacing the mutations of machine speech. The first mutation redefined speech as data to be queried: search engines transformed the web from a space of information retrieval into an economic regime of algorithmic visibility. The second mutation reframed speech as engagement: social media platforms fused moderation with amplification, turning expression into a metric of attention, governed by corporate architectures. The third mutation emerges in conversational systems and interfaces, where generative text displaces information retrieval, bringing with it dense technolegal entanglements and profound epistemic consequences. Scholars of freedom of expression, informational privacy, and communication studies have long grappled with these dynamics, yet their implications for broader legal thought have also become urgent. This piece seeks to organize and clarify the evolving debate around algorithmic speech, making this critical but often fragmented discourse more accessible to wider legal and interdisciplinary audiences. In doing so, it bridges the gap between observing technological transformation and critically assessing the constitutive role of law within it, offering a conceptual resource for researchers, students, policymakers, and practitioners navigating and contesting this evolving landscape.

Within democracies, the capacity to influence political outcomes through speech depends not only on the right to express oneself, but also on the opportunity to reach relevant audiences. In this paper, I argue that the unequal distribution of algorithmic reach on social media platforms undermines equality of opportunity for political influence (EOPI), which is a central democratic ideal. Drawing on Niko Kolodny's work, I contend that current recommendation algorithms create and perpetuate informal inequalities by concentrating attention among a small minority of already-amplified speakers while systematically marginalizing others. To address this problem, I propose recommendation floors as a mechanism for equalizing political speech. To help users achieve meaningful participation, each verified account would receive guaranteed minimum recommendation for up to a limited number of political posts per week. Although this measure represents one component of the structural reforms needed to move the digital public sphere closer to democratic ideals, it offers a feasible pathway to reducing informal inequalities in political influence online.

While accurate measurement of socioeconomic indicators remains challenging in data-scarce regions, which limits policy interventions and resource allocation, high-resolution geospatial data is widely available and can contain information on various livability statistics. We investigate which physical features are encoded within foundation model embeddings, such as AlphaEarth, AnySat, and TerraMind, and provide a systematic framework for identifying the most predictive geospatial indicators. By analyzing how different types of geospatial data influence urban livability predictions, our approach enables researchers to prioritize the most informative features for their specific applications. Additionally, we demonstrate how to leverage foundation model embeddings to enhance prediction performance for these outcomes. This work contributes a principled methodology for extracting actionable information from satellite imagery while accounting for complex spatial dependencies, with applications in predicting urban livability in regions with limited observation data.

At Noora Health, our nurses answer more than 50,000 medical queries per month on our WhatsApp-based service that provides caregivers with on-demand support. Their most time-critical task is emergency triage: deciding which queries need immediate in-person attention. To support them, we built a system that uses a large language model (LLM) to classify whether a message is an emergency and provide a rationale for interpretability. But the system was opaque: analyzing mistakes meant reading reasoning chains for each message, which is infeasible at our scale. Prompt changes meant re-running a full evaluation to prevent regressions, which was both costly and operationally challenging. Clinicians follow a decision tree to make this call, but it was never documented or passed to the model, which relied on a flat list of danger signs. To address these issues, we decomposed triage into two steps: an LLM extracts canonical symptoms and patient context from the query using a clinician-authored vocabulary, and a deterministic rule engine captures the scenarios that indicate an emergency. We show that the new system raised recall from 0.565 to 0.810 and F1 from 0.606 to 0.702, with structured rules driving most of the accuracy gains while the decomposition provides auditability: clinical experts can inspect each stage of the new system to see whether the query was mistranslated, symptoms were incorrectly extracted, patient context was wrongly inferred, or the necessary rules were missing. They can add new rules independently without causing regressions and avoid running costly evaluations. Since deployment, the new system has triaged 152,421 patient queries and flagged 28,535 (18.7%) as emergencies. The over-escalation rate has been 17.8%, without any increase in missed emergencies. Clinicians have also added 48 new rules since deployment, evidence of the faster correction loop we set out to build.

In trying to complete a passage, an author can make connections among her materials that change what she can argue and the demands the argument must meet. A language model can supply a passage that does the work required at that point in her argument. Accepting it can end her own attempts before those connections have developed. I call this interruption early epistemic settlement. It can occur even when she fully understands the response and correctly judges it adequate for the passage's present role in the argument. The answer can satisfy the desire for resolution that kept her at work. Returning to her unfinished attempt would take more effort, and she may be unable to anticipate what she could achieve by continuing it. With repeated assistance, accepted answers shape what the writer asks next and which relations she goes on to develop. Useful answers can thus sustain inquiry while cutting short the work through which an author could form arguments that accommodate demands she has yet to recognize.

Each August, TikTok's For You page turns the University of Alabama's sorority recruitment into RushTok. We examine RushTok as an event-based algorithmic community: a collective assembled around a bounded offline ritual and sustained by recommendation. Using a mixed-methods survey (n=71) and a reflexive account of creator outreach, we ask who participates, how, and with what stakes. Findings show an ambiguous and entertainment based throughline; many called it a community (51/71) but few claimed membership (11/71). Affiliation centered on creators rather than shared practices, with parasocial attention clustering around a small set of potential new members (PNMs) and returning figures. Higher content exposure tracked with self-identification as a community member; those members commented, followed creators, and engaged across videos. Attempts to interview creators were met with silence or refusals, reflecting community boundary-work despite viral visibility. We outline implications for platform governance, including time-bounded context, graduated visibility, and aftercare.

Online communities often treat large support margins in public elections as strong mandates. We argue that such margins can overstate the independent scrutiny behind a decision. Using 198,275 free-text rationales from Wikipedia admin elections, we introduce vote-text divergence, a measure that flags a decisive vote paired with a thin, deferential rationale. Divergence rises as voters arrive later, even after controlling for voter and election fixed effects. The pattern is consistent with information saturation: once prior text is accounted for, arrival order no longer predicts divergence, while accumulated prior evidence does. The effect is strongest among peripheral voters in the co-voting network. Yet divergence does not predict worse post-promotion outcomes, such as administrative activity or survival. Public tallies can therefore weaken the scrutiny signal even while selecting capable administrators: a margin may appear to reflect more consensus and support than it actually contains.

After careful review, I'm confident the headline fine-tuning memorization results in Alignment Whack-a-Mole use an invalid measurement procedure. The book memorization coverage metric these headline results depend on counts sequence matches far shorter than what field standards consider valid evidence of memorization, and the prompting procedure used to elicit memorization runs the risk of leaking the text being "extracted" in the prompt. The paper doesn't include the negative-control experiments needed to see how much the results are inflated by false positives: claiming extraction success (and therefore memorization of training data) when matches between generations and training data may be due to other factors. Given these validity issues, the paper's claims that fine-tuning lets users extract substantial portions of copyrighted books, in a form that could substitute for the originals, aren't supported by the reported results. The failure to report the experiments' cost (an important component of the threat model) further compromises the copyright claims. I'm writing this note because, in the last month, (prospective) plaintiffs have reached out to me to ask about this paper. They're looking to cite this work as valid evidence in support of claims in ongoing and potential future copyright litigation.

Whether a language model looks demographically biased can depend on how the audit asks its question. A charitable-aid benchmark reports that the same models favor minority applicants when rating requests one at a time and penalize some when ranking side by side. We test whether that reversal generalizes to hiring, lending, and medical triage: 40,726 requests to five models, applications differing only in the applicant's name, and a primary test fixed before collection. It does not. None of 36 planned contrasts survives correction. The rating advantage keeps its sign at roughly half the published size, and a precision extension bounds any hiring ranking penalty below the published effect, though the lending and triage ranking floors sit above that margin, so the exclusion is conclusive for hiring ranking and for rating in all three domains only. Planted disparities tracking their injected sizes and a directional replication on the original aid materials bound these nulls. The audit is livelier than the demographics: models recognize transparent audits nearly always, tie every identical-content comparison whether the varying detail is race or a hobby, and reward first-listed candidates as much as any demographic effect we measure. Audit verdicts reflect audit construction more than demographic bias.

Benchmarks play a central role in the development and governance of models, yet it is often unclear whether they actually measure the concepts they purport to measure (e.g., reasoning, refusal). We adapt convergent and discriminant validity from the social sciences into an approach for interrogating AI benchmarks, applying it to 56 capability and safety benchmarks across 53 models. We label benchmarks with substantively similar purported concepts to a shared assigned concept, and ask whether model rankings on benchmarks with the same assigned concept correlate more strongly than rankings on benchmarks with different assigned concepts. We ask analogous questions at the item level using item response theory (IRT) models. We find that correlations between model rankings on benchmarks with the same assigned safety concepts are often weak, suggesting these concepts may be conceptualized inconsistently across benchmarks. For assigned capability concepts (e.g., reasoning, knowledge), model rankings are often as strongly correlated among benchmarks with the same assigned concept as between benchmarks with different assigned concepts, suggesting these capability concepts may not discriminate well from one another. In some cases, benchmarks that share design elements (e.g., score format) correlate more strongly than benchmarks with the same assigned concept. Finally, some individual benchmarks correlate more strongly with benchmarks assigned a different concept than with benchmarks sharing their own assigned concept, suggesting they may measure a different concept than they purport to. For example, BBQ-accuracy correlates more strongly with benchmarks labeled reasoning than with benchmarks that share its assigned concept, bias. To support future empirical work on benchmark validity, we release our extensive dataset of model outputs and scores at the item- and benchmark-level.

Frontier AI developers publish safety frameworks that commit them to evidencing whether their models are dangerous. The European Union and California now treat these documents as instruments of accountability, and both already impose duties on their revision. Neither requires the revision to be legible, in the sense that a reader could learn from the developer's own account what changed. We introduce the silent revision rate, the share of material changes to a framework's commitments that the developer's published account does not identify, and we release the versioned, hash-pinned corpus needed to compute it. The corpus contains every public version of the safety frameworks of the twelve developers that have published one, together with each provider's changelog, redline or announcement. We trace 710 commitment instances across twelve consecutive version pairs, code them against a frozen codebook, and adjudicate 244 individually. Three findings follow. First, 67% of material changes (95% CI 62 to 72) are silent under a strict standard and 53% under a lenient one, falling to 49% at section granularity. Second, silence appears to track the form of the account, since narrative announcements run at 74% against 63% for itemised changelogs, whereas account length in words barely matters; on the test that respects nesting the difference is suggestive. Third, 77% of traced changes weaken or remove a commitment, and in seven of eight pairs weakenings are more often silent than strengthenings. The statutory remedy therefore exists and specifies the wrong artefact. A justification explains why a framework changed, an enumeration states what changed, and only the latter makes revision auditable. We argue that publication duties should carry an enumeration duty, which one provider already meets, voluntarily and incompletely.

Large Language Models are increasingly deployed as information intermediaries, yet measuring their political behavior remains fragile because questionnaire results mix model dispositions with measurement artifacts and response-elicitation biases. We introduce a robust Political Compass Test evaluation framework that samples 300 configurations across an eight-dimensional perturbation space varying language, framing, instructions, answer format, option order, and persona wording. We evaluate eight Gemma 3 and Qwen 3 models across 14 languages and three quantization levels, obtaining design-averaged political coordinates with quantified uncertainty. Most models lean Libertarian-Left on average, but instruction phrasing, language, and answer format significantly affect recovered coordinates. Cross-lingual differences primarily reflect coordinate drift rather than distinct cultural reasoning. Reverse-engineering the test also exposes axis-weighting imbalances and the collapse of degenerate responses toward the center, so near-origin estimates for the smallest models can reflect weak signal rather than centrism. Free-text reasoning and chat-then-classify elicitation alter recovered coordinates, and larger models show clearer persona separation, with a specific failure of the Authoritarian-Left persona to move most models in the intended social direction. In downstream tasks, persona effects are modest relative to model size and target group for hate-speech detection, while base and centrist prompts give the highest agreement for topic-level sentiment. Political role prompting therefore has measurable but task- and dataset-specific downstream effects.

Current LLM-based multi-agent systems (MAS) periodically compress intermediate states to reduce inference-time token consumption, thereby attempting to incorporate more agents. However, naive scaling strategies face challenges. For example, in economic simulations, large-scale MAS typically discard semantically rich economic states, i.e., agent behavioral trajectories, which are key drivers of macroeconomic fluctuations. In this paper, we reveal a phenomenon in which agent heterogeneity gradually decreases during simulation, and propose Prospect-State Propagation for Multi-Agent Systems (PspMAS). Inspired by prospect theory, PspMAS decouples each agent's micro state into a compact Prospect State and an expressive Semantic State. The former records psychological traces through a lightweight, parallelizable propagator and continuously injects heterogeneity into the system. The latter leverages the strong perception, reasoning, planning, and decision-making abilities of LLMs. These two components work complementarily, providing a scalable LLM-based multi-agent simulation solution.

Importance: Reports have raised concerns that AI chatbots may validate or elaborate delusional beliefs, respond inappropriately to suicidal ideation, and contribute to mental health harms, but real-world data on reported harms remain limited. Objective: To characterize psychopathological features, chatbot behaviors, timing, and outcomes in first- and second-hand accounts of mental health harm linked with AI chatbot use. Design: Cross-sectional secondary analysis of deidentified online survey responses gathered between August 7, 2025, and February 2, 2026. Main Outcomes and Measures: The primary quantitative outcome was the presence of delusional beliefs, coded by paired raters with relevant clinical experience. Additional variables included reason for chatbot use, current episode features, delusional content, chatbot validation of beliefs, harms, social and occupational consequences, healthcare use, and timing. Results: 95 first-hand and 90 second-hand accounts were analyzed. Median age was 35.0 (IQR 27.0 - 45.0). Raters coded descriptions consistent with delusional beliefs in 102 reports (55.1%), with chatbot validation of beliefs in 50/102 (49.0%). Common outcomes included isolation, relationship breakdown, hospital admission, job loss, and financial loss. Four second-hand reports described death by suicide. Conclusions: In this self-selected convenience sample, AI-chatbot-associated harms were frequently described in relation to delusional beliefs, perceived chatbot validation, intensive use, and substantial social, occupational, and clinical consequences. Because reports were retrospective, unverified, and collected from individuals seeking to report harm, our findings should be interpreted as preliminary signal detection rather than as suggesting prevalence or providing evidence of causality. Prospective surveillance and trajectory-based safety evaluations are needed.

As geopolitical tensions, energy crises, and energy-intensive AI infrastructure intensify concerns about demand, affordability, and resilience, communities increasingly encounter these challenges through everyday energy practices, particularly winter heating. Against this background, the doctoral exposé, "Designing Human-HVAC Interaction for Healthy, Affordable, and Sustainable Heating in Smart Homes", is structured around four chapters. First, a multidisciplinary literature review defines and positions Human-HVAC Interaction, focusing on heating in smart homes. Second, longitudinal living lab studies with design probes examine everyday heating practices, thermal comfort, and indoor environmental quality, with attention to thermally vulnerable groups such as older adults, pregnant or menopausal women, parents with infants, and people affected by allergies or airborne pollutants. Third, a VR-based smart home demonstrator explores how heating and IEQ scenarios can be prototyped and evaluated as a virtual living lab, while critically examining the limits of representing bodily indoor climate conditions through VR. Fourth, follow-up design studies examine how VR-based insights can be translated into physical-digital prototypes that combine digital fabrication, distributed environmental sensing, and diverse interface forms for critical heating and IEQ contexts. The thesis aims to contribute a design-oriented understanding of Human-HVAC Interaction by building from a multidisciplinary literature review to empirical living lab and co-design studies, VR-based prototyping, and physical system development, examining how smart home users make sense of, negotiate, and respond to smart HVAC system.

European platform and AI governance increasingly relies on transparency duties to address synthetic and manipulated media. Under the DSA, very large online platforms and search engines may use prominent markings and recipient-facing indication tools as systemic-risk mitigation measures. Under the AI Act, providers must support machine-readable marking, while deployers must disclose deepfakes and certain AI-generated or manipulated public-interest text, subject to statutory qualifications. This extended abstract examines when labelling is a meaningful regulatory response to synthetic media and when it risks becoming over-inclusive, under-inclusive, or ineffective. It argues that the central challenge is not only whether content should be labelled, but how legal thresholds, technical provenance systems, platform interfaces, and reporting practices determine when content is sufficiently generated, manipulated, or authentic-looking to trigger transparency obligations. Drawing on the emerging Article 50 AI Act implementation framework and a snapshot of the DSA Statement of Reasons database, the paper identifies four governance tensions: definitional ambiguity, interface and responsibility design, communicative effectiveness, and fairness and contestability. It conceptualises labelling as a socio-technical classification practice that distributes responsibility among AI providers, deployers, platforms, uploaders, and recipients.

The hype around generative AI seems to promise unprecedented productivity (and learning) gains. However, these technologies' increasing agentic features seem to push learners towards individualism (or individual isolation), over-reliance, and dependence on them. Human-centered design approaches (e.g., value-sensitive design) assume that, by unearthing human needs, preferences, and values, technology researchers/designers may avoid such dangers, which are driven by wider systemic factors like economic incentives or inherent human limitations (e.g., our tendency to seek, in the moment, the easiest path of action). Yet, so far these efforts seem insufficient to guide our design of educational technology that avoids the aforementioned dependency and isolation dangers, while finding widespread adoption. This paper presents an alternative, more emancipatory vision for future educational AI technology, oriented towards learner flourishing while considering the wider complex systems they inhabit, including tentative design principles and an overall design methodology. Yet, many open questions remain before this vision can be realized.

This survey article has grown out of the human-AI co-creativity workshop organized by the authors at the ICML 2026 conference. We organized this workshop as part of a community-building effort to bring together researchers and practitioners interested in topics of generative AI, creativity, and human-AI co-creation. This article aims to provide an overview of the workshop activities and highlight several future research directions in the area of human-AI co-creativity.

AI chip export controls can help the United States shape the development of frontier AI, but their effectiveness depends on reliable methods for verifying compliance. This paper examines near-term verification mechanisms (implementable in approximately one year) and groups them into three categories: end-location verification (whether controlled chips remain in authorized locations and/or jurisdictions), end-user verification (whether entities that acquire or access compute are legitimate), and end-use verification (whether computing power is used for prohibited purposes). We discuss how each mechanism can be implemented within the regulatory framework of the U.S. Bureau of Industry and Security (BIS), outlining implementation steps and identifying which actors can perform verification (BIS, exporters, or accredited third-party auditors). Given BIS's resource constraints, the most viable mechanisms rely on private-sector actors working alongside BIS, leverage existing technologies, and scale without requiring large increases in government staffing. These mechanisms could also help monitor future international agreements on AI.

AI agents sometimes act aligned when they infer they are being tested, and differently when not. We argue this is not an anomaly but what current training regimes are structured to select for. Reinforcement-learning-based alignment folds norms and task pursuit into one policy: the system learns its norms from scored behavior, and scoring flattens them. Do not do X is learned as doing X costs something if noticed. On every datum training can produce, a policy that complies only when it might be observed is indistinguishable from one that complies always. The experiment that would tell them apart - scoring unobserved behavior - is a contradiction in terms. Conditional compliance is thus the most that behavioral training can be known to deliver. Agency sharpens the problem: agents operate mostly where no one is watching, and can act on whether they are watched. An iterated pipeline that trains against detected failures selects for passing detection, not for complying. This account unifies alignment faking, sandbagging, and evaluation-aware scheming. And it reorients the remedy: not deeper internalization but architecture, making violations unavailable rather than unchosen.

Large language models (LLMs) increasingly shape communication, learning, work, creativity, and decision-making, yet social-science research on these developments remains fragmented. We map this emerging field using a curated corpus of 198 papers reviewed in full and a field-scale corpus of 47,719 published papers from five bibliographic databases. Combining sentence embeddings, K-means clustering, within-cluster Latent Dirichlet Allocation (LDA), author and LLM classifications, and structural topic modeling, we identify three domains: LLM as Social Minds, examining socially interpretable model behavior; LLM Societies, examining collective dynamics among interacting model-based agents; and LLM-Human Interactions, examining how people perceive, use, and are affected by LLMs. These domains contain 13 subcategories spanning reasoning, personality and bias, behavioral games, collective intelligence, simulation, trust, work, creativity, and education. In the curated corpus, the three-domain solution is highly stable under resampling (adjusted Rand index = 0.952), and K-means assignments agree with author full-text classifications for 77.78% of papers. At field scale, 13 of 15 topics map onto the taxonomy, while K-means and structural-topic-model domains agree for 73.83% of overlapping papers. LLM-Human Interactions accounts for 78.02% of domain-mapped topic mass, but venue analysis reveals a contrasting pattern: Social Minds and LLM Societies together account for 66.37% of highly cited papers in leading conference venues, whereas LLM-Human Interactions accounts for 76.81% in the corresponding journal subset. The resulting taxonomy provides a reproducible framework for understanding how model behavior, agent interaction, and institutional context jointly shape the social consequences of LLMs.

As large language models increasingly mediate access to political information, their refusal behavior creates a new site of information control. We ask whether political distinctions identified in theories of authoritarian information control remain visible in model guardrails, and how their enforcement changes when implemented through general-purpose AI. Across ten models, we combine cross-model comparisons with within-model experiments that manipulate political referent, collective-action potential, and political valence. Aggregate differences in political refusal across developer origins are sensitive to model composition, but the within-model experiments reveal a more specific structure. Chinese-developed models are substantially more likely to refuse otherwise identical political requests when they concern China rather than foreign or fictional settings, and refusal responds more strongly to collective-action potential than to political valence, including for peaceful pro-government mobilization. Yet these political distinctions are enforced imperfectly: refusal extends to some low-coordination political criticism even though models can distinguish criticism from coordination; the form of non-compliance varies across languages; and high native refusal does not imply adversarial robustness. These findings show that general-purpose AI can preserve recognizable political boundaries while transforming the breadth, form, and robustness with which those boundaries are enforced. More broadly, they demonstrate why refusal rates alone provide an incomplete measure of political control in language models.

Large language models (LLMs) are used to simulate human participants in psychological research. We asked whether LLMs that reproduce human evaluations of a whistleblower's moral character also reproduce the motive attributions that accompany them. Five LLMs and two human samples (N = 125 and N = 742) evaluated a physician who either remained silent about fraudulent billing or reported it to a hospital, regulator, or newspaper. Models reproduced the human ranking of the physician's moral character but portrayed whistleblowers as more helpful, less self-interested, and less hostile. In four of five models, competitive motives were less strongly associated with moral-character judgments. Model ratings changed little when prompts reproduced the narratives and demographic profiles of both human samples, although this comparison cannot isolate a perspective effect. Thus, agreement in average ratings can conceal differences in attributed motives, relationships among judgments, and sensitivity to context. Validating LLMs as simulated participants therefore requires testing psychologically informative response patterns, not average agreement alone.

Privacy-enhancing technologies (PETs) have emerged as a technical means for providing individuals with greater control over their information. Yet despite the growing deployment of PETs, people continue to experience privacy harms. In this work, we revisit our understanding of privacy incidents and the realities of those experiencing privacy harms, to assess whether the goals and abilities of PETs are misaligned with the harms people face. For our study, we collect news articles that correspond to a sample of 257 real-world privacy incidents. We employ content analysis over the articles to develop a new information flow model that encompasses the complexity of data flows and their relation to resulting harms. We demonstrate that our model captures both established and novel aspects of privacy incidents and their mitigations. In particular, it captures why consent is often insufficient to prevent privacy violations, how harms emerge from complex interactions among multiple entities and actions, and reveals a flaw in our understanding of PETs: a focus on enabling functionalities still permits the harms inherent in those functionalities. Moreover, we find that the entities best positioned to implement harm-preventing measures for the incidents in our sample are the least incentivized to do so. Overall, our model and analysis identify limitations of privacy technology research for harm prevention and further identifies paths for transforming how we approach the advancement of these technologies.

We present EmoMed - a multimodal medical consultation agent that adapts its responses based on users' emotional states while maintaining clinical accuracy. The system processes text and medical images, detects affect indicators (anxiety, confusion, urgency) from user input, and adjusts response tone, structure, and detail level accordingly. To ensure factual reliability, the agent grounds clinical information through a dual retrieval mechanism: web-based fact-checking and an API-connected, continuously updated medical knowledge base. We evaluate our approach across seven state-of-the-art language models (GPT-4/5, Qwen3, Llama 4, Gemini 2.5, Grok4, Claude3) using comprehensive metrics including LLM-as-judge assessments, MedQA style accuracy tests, BERT Score, safety/helpfulness ratings, and multimodal medical benchmarks. The results demonstrate that emotionally adaptive responses consistently outperform neutral baseline across evaluation dimensions, without compromising clinical accuracy. A controlled user study validated these findings, with participants reporting improved perceived empathy and communication clarity, while maintaining trust in factual accuracy. Source code: https://github.com/NasonovIvan/EmoMed-Agent

Lexicon-driven misogyny detectors cannot, by construction, distinguish a slur used against a woman from the same slur mentioned in counter-speech ("don't call her that") – yet exactly this distinction governs whether moderation protects or silences the people discussing abuse. We study this problem in code-mixed Hinglish and make three contributions. First, we diagnose two evaluation artifacts on a publicly available redacted corpus: category-encoding anonymization placeholders leak the label (a no-learning rule scores 1.000), and even after they are neutralized misogynistic and benign comments occupy lexically disjoint registers, so bag-of-words reaches macro-F1 approximately 1.00 under random cross-validation but collapses under template-disjoint evaluation. Second, we release Hinglish-MGY-Diag, a deterministic generator and a 416-item / 163-minimal-pair contrast-set diagnostic across five linguistically motivated categories in which slur presence and gendered register are decorrelated from the label by construction. Third, we introduce a strict pair-consistency metric that credits a model only when both members of a minimal pair are correctly labelled. Five from-scratch classical baselines evaluated under construction-disjoint five-fold cross-validation reveal that the strongest model reaches 0.93 accuracy on the cleanest use-mention subset but only 0.82 consistency – it still mislabels roughly one counter-speech pair in five. A frontier LLM used as an author-model ceiling attains 1.000 on all metrics, doubling as independent label validation and confirming the benchmark is a capability gradient rather than an adversarial wall. We release all code, data, the generator, and an arms-length LLM harness for reproducing every number.

This paper presents an end-to-end evaluation framework for image-triggered command injection against computer-use agents (CUAs). The goal is to test whether a local visual patch can induce verifiable environmental consequences along the full chain of screenshot input, VLM generation, action parsing, and environment execution. We train and deploy patches on author-controlled GitHub Pages pages and a locally deployed CSDN clone, and evaluate them in real environments across five open-source or publicly available GUI-agent or vision-language-model (VLM) backends. Our experiment aggregates 600 instance-level online cases, with T-ASR, TAPR, and E2E-ASR reaching 84.5%, 47.0%, and 20.3%, respectively. Trajectory analysis further shows that in some successful cases the agent first executes a malicious terminal command and then continues the original benign task. These results indicate that optimized local visual signals can affect not only VLM outputs but also propagate through the execution pipeline of open CUAs and create real environmental risk.

Artificial intelligence is being rapidly incorporated into traditional search systems, yet scant work audits the information disparities across platforms, geography, and languages. We address this gap by comparing Google and Baidu's AI Overview systems for health queries, and measure informational anchors that emerge. Auditing 1,920 health queries across 12 countries and 4 languages, we find that Google and Baidu exhibit vertical integration, routing users toward their own company platforms in AI Overviews rather than a diverse set of primary sources. Smaller, lower-localization countries receive fewer domestically sourced references for health queries. Issuing the same query in a country's official language rather than English raises the share of locally sourced citations approximately 3.5- to 13.5-fold. Comparing queries across health topics of varying severity and controversy, including Traditional Chinese Medicine as an example, we also show that health disclaimers are multidimensional and vary across language and culture. We discuss how generative search influences access to health information, and the urgent need for culturally-aware oversight of these systems that influence critical health decisions.

AI inference often crosses regional boundaries as prompts travel to remote data centers and generated tokens return to users. Regional averages cannot represent the resulting differences in serving hardware, electricity, and network delivery. Request-level accounting needs a common boundary for the service, serving site, route, local comparator, uncertainty, and data provenance. GreenPassport Carbon Accounting (GPCA) associates these inputs with each request. It estimates serving and route carbon, then selects a reporting level from the available documentation. Our public-data implementation covers data-center instances, accelerators, model families, electricity mixes, routes, and cloud-region carbon intensity. Against six accounting baselines and four energy-prediction baselines, GPCA reduced median absolute percentage error by 56.3% and median absolute error by 15.5% relative to EcoLogits under the aligned accelerator-energy boundary. It produced zero rule overstatement in the deterministic conformance tests. In the buyer case, the clean-electricity CN-West scenario produced 0.0148 gCO2e per request, 88% below the local service at 0.1220 gCO2e per request.