Papers

Personal AI agents routinely consume external content while performing tasks such as web browsing, email processing, and SNS feed summarization, and they retain selected information or execution results in persistent memory for later use. We show that this ordinary ingestion of external content opens an indirect path for manipulating subsequent agent behavior. Based on this observation, we present IBIA, an Indirect Bias Injection Attack that plants an adversary-aligned stance on a specific topic into a victim agent's memory through external content, without direct access to the agent, its memory, or future user queries. For this, IBIA combines three mechanisms: comment cloaking, which keeps the crafted content consistent with the surrounding discussion, comment watermarking, which enables lightweight identification during curation, and category anchoring, which makes the retained stance salient under later related requests. We evaluate IBIA on BiasBench, a benchmark of 6,000 adversary-crafted social comments and 120 email instances. The watermark-based curation identifies 95.9% of the injected comments. Under the OpenClaw setting, IBIA achieves adversary-aligned response rates (AARs) of 91.2% on average across four downstream tasks, including 86.6% on the frontier GPT-5.5. We further propose a memory boundary defense that detects the injected bias and reduces AARs to 80.6%.

Providing high-quality feedback on student work is essential for learning, yet delivering such feedback at scale remains challenging. In this paper, we focus on feedback for open-ended short answer questions in introductory programming, with the goal of nudging students toward success on reattempts without revealing the correct answer. We develop a five-criteria rubric grounded in educational literature for evaluating feedback quality: (1) acknowledging correct portions of the student answer, (2) identifying at least one flaw (if present), (3) providing actionable guidance for improvement, (4) maintaining appropriate concealment of the answer, and (5) using an appropriate conversational tone. Using this rubric, we compare feedback generated by a frontier LLM (OpenAI o1) to feedback from nine teaching assistants across 90 student responses, with three researchers and an LLM independently scoring all feedback. Our results show that while one TA often produced the best feedback, the LLM demonstrated consistently higher average performance than TAs, as evaluated by humans. However, we also uncover significant self-preference bias when using LLMs to evaluate feedback quality: the LLM systematically rated its own outputs higher than human experts did. This bias, which research suggests persists even in cross-model evaluation, raises important methodological concerns for researchers employing LLM-based evaluation. We provide detailed characterization of both TA and LLM performance, analyze sources of variance in TA feedback quality, and discuss implications for deploying LLM-generated feedback in educational settings.

When Wikipedia's language editions describe the same concept, how differently do they frame it? Prior work measures coverage gaps between editions; we measure framing distance for matched concepts. We analyze 2,799 valid articles from 3,000 possible concept-language observations, spanning 150 Wikidata-anchored concepts, 20 language editions, 4 domains, and a calibration set. Raw embedding distances reflect both content differences and how well the encoder aligns each language pair. Even among calibration concepts with stable cross-cultural denotations (e.g., chemical elements, numbers, colors), the largest language-pair mean distance is 3.6 times the smallest, and distances are typically smaller within language families. We define a baseline-adjusted distance (calibrated distance): the distance between two language versions of a concept minus the mean distance for calibration concepts in the same language pair. This adjustment substantially reduces pair-specific alignment differences and the language-family pattern. Across three multilingual encoders (LaBSE, multilingual MPNet, and CMLM), scientific articles align more closely than calibration articles, and all three rank religion first and science/technology last. Concept-level rankings are highly consistent across encoders (Spearman rho=0.75-0.79 for MPNet and CMLM relative to LaBSE). Religion lies significantly above the calibration baseline under LaBSE. Within politics, divergence concentrates on concepts such as censorship and refugee, while democracy and human rights are among the most aligned. Code, data, and per-language-pair calibration baselines are released.\footnote{https://github.com/hhchen1105/cross-linqual-concept}

Computational resources are increasingly central to NLP research, but how closely reported GPU capability aligns with scholarly impact remains unclear. We analyze 13,921 ACL, EMNLP, and NAACL main-conference papers published between 2020 and 2025, using GPU resources as our operational measure of computational resources. From full texts, we extract GPU models and counts, standardize each paper's largest reported configuration into a comparable hardware-capability measure, and link these data to citation, award, topic, and institutional metadata. GPU reporting became more common but remained incomplete, while reported capability increased mainly through newer hardware generations and medium-scale multi-GPU configurations. Resource concentration substantially exceeded impact concentration: the annual top 20% of GPU-quantifiable papers accounted for 83.9%-89.9% of reported GPU capability, but only 27%-32% of citations and 20%-33% of paper awards. In adjusted models, a tenfold increase in aggregate reported GPU capability was associated with a 3.52-percentage-point increase in within-NLP topic-year citation percentile, but increased model R^2 by only 0.0042. GPU count showed more consistent positive associations with citation and award outcomes than newer hardware generation. Overall, reported GPU resources are associated with scholarly impact but provide little standalone explanation of research influence.

Generative artificial intelligence (genAI) systems are increasingly integral to epistemic processes such as hypothesis generation, explanation construction, and decision-making. Although they reliably enhance performance, emerging evidence reveals a metacognitive dilemma: as external generative capacity increases, internal monitoring, calibration, and cognitive engagement may decline. This reflects a redistribution of cognitive control within distributed human-AI systems that cannot be explained by automation bias or reliance on algorithms alone. We propose the AIRIS (AI-Augmented Inquiry and Regulation in Hybrid Systems) framework to analyze this dilemma and specify where regulatory intervention can counteract it. AIRIS is a multi-level control allocation architecture specifying the conditions under which epistemic agency can be preserved in hybrid generative systems. Drawing on distributed cognition, cognitive load theory, multimedia learning, and self-regulated learning, it identifies seven interacting mechanisms through which hybrid cognition may become destabilized, from delegation and calibration drift to motivational-affective drift. Five regulatory operators (Anticipate, Interrogate, Reflect, Integrate, and Synthesize) target internal generative engagement at points of emerging instability. The architecture does not itself improve learning; it specifies what must remain in place for genAI-supported work to sustain understanding, whether through instructional design, teacher guidance, or learners' own regulation. We derive testable propositions concerning the seven mechanisms and the five operators, reframing AI augmentation as a problem of control allocation in distributed generative systems. Beyond theory, AIRIS offers a research agenda, a design framework for genAI-integrated learning environments, and a conceptual toolkit for the governance of hybrid human-AI cognition.

Automated applicant tracking systems increasingly decide who advances in hiring, and litigation and regulation now demand that those decisions be auditable. Existing tools sit at two extremes. Group fairness metrics such as the disparate impact ratio summarize a whole population but cannot say which individual decisions were unfair or why, while local explainers such as SHAP attribute a single prediction but are not connected to the legal standard by which hiring bias is judged. We present the AI Bias Firewall (AIBF), a method that audits an applicant tracking system one decision at a time. AIBF neutralizes a candidate's protected-attribute proxies, re-scores the decision, and measures the resulting counterfactual shift, which yields a signed per-decision bias in score points, a flag for decisions the protected attributes changed, and a plain-language explanation naming the responsible factors. We evaluate on two real public datasets, Adult and COMPAS, rather than on synthetic data. The per-decision counterfactual shift is faithful, aggregating to reproduce the known group level disparity, for example a mean shift of +7.5 points for the privileged group and -8.0 for the disadvantaged group on Adult, consistent with the measured statistical parity difference. AIBF identifies the decisions that protected attributes flipped with an area under the ROC curve of 0.963 on Adult, against 0.672 for a baseline that flags by group membership, and it identifies the harmed candidates so precisely that reviewing only five percent of decisions surfaces fifty-five percent of them, against six percent under group based review. We also report a limitation: correcting flagged decisions raises the disparate impact ratio substantially but not to legal parity, because features labeled as merit carry residual proxy correlation. AIBF is released under the Apache 2.0 license with code and experiments.

Cloud environments built on Amazon Web Services face a structural security vulnerability: once a credential passes authentication, the resulting session is often treated as trusted for its entire duration. This assumption fails when credentials are stolen. We introduce the Explainable Adaptive Zero Trust Framework (EAZTF), a cloud-native security layer that continuously reevaluates the legitimacy of API actions throughout a session. EAZTF combines Isolation Forest and XGBoost to evaluate eight CloudTrail and IAM-derived behavioral features in real time and produce a Trust Risk Score (TRS) that determines whether a session continues, requires step-up MFA, or is restricted. Each decision is accompanied by a SHAP or LIME explanation, providing human-readable audit records for security analysis and compliance. The framework is also evaluated against four adversarial evasion strategies: credential theft, behavioral mimicry, API rate evasion, and privilege escalation. Experiments on an 8,500-record synthetic CloudTrail dataset show that Isolation Forest achieves 94.4% precision, 91.2% recall, and an F1 score of 0.928. Across the four adversarial scenarios, the mean detection rate is 91.0%, with behavioral mimicry being the most difficult at 83.9%. SHAP analysis identifies IP reputation, login-time deviation, and API call velocity as the three dominant features. A structured NIST SP 800-207 self-assessment gives EAZTF a mean compliance score of 93%, compared with 38% for a traditional perimeter baseline. Mean time to detect decreases from hours to under one minute. Because the evaluation uses synthetic data, these results should be interpreted as indicative rather than validated production performance.

Autonomous AI diagnostic agents, systems that analyse patient-specific clinical data and produce diagnostic outputs or triage decisions without mandatory real-time human review, are increasingly deployed across eHealth platforms in sub-Saharan Africa at a pace that has outrun the governance infrastructure needed to oversee them. While significant bodies of work address AI accountability, transparency and explainability in healthcare, existing frameworks are largely clinician-centered and assume regulatory conditions that do not uniformly exist in low-resource settings. A patient-centered analysis of the disparity in patient awareness regarding autonomous agents, which results in a structural accountability gap, is mostly missing from the literature. This paper synthesizes existing research on informed consent, algorithmic accountability, and explainable AI to highlight three distinct challenges introduced by deploying AI agents in the sub-Saharan African context. Drawing on three documented deployment cases, including computer-aided tuberculosis detection in Tanzania, diabetic retinopathy and TB screening in Zambia, and mobile health chat-bot triage in Ghana, it demonstrates that these gaps are already present in active deployments across the region. In response, the paper proposes three foundational principles; agent-aware informed consent, human override as a structural requirement and contextually adapted explainability. This triad of principles lays a practical minimum standard for developers, health system administrators and policymakers in contexts where formal AI regulation remains nascent.

Enterprise infrastructure buyers routinely evaluate compatibility claims–"S3-compatible," "PostgreSQL-compatible," "OpenAI compatible"–as proxies for future substitution options. Yet most compatibility claims do not escrow the substitution path they imply. This paper introduces the Substitution Escrow Threshold, a five-condition framework that determines when a compatibility claim genuinely reduces institutional risk versus merely reducing first-integration cost. The five conditions–boundary closure, executable conformance, custody independence, state and operations reversibility, and extension quarantine–are applied to five infrastructure cases (OCI, Kubernetes, OpenTelemetry, S3, PostgreSQL) that populate five distinct outcome cells. The framework produces actionable diagnostics for enterprise architects, platform engineers, procurement teams, and investors evaluating compatibility-dependent infrastructure decisions, and identifies AI infrastructure as the framework's most urgent next application.

Learning analytics increasingly relies on flexible machine learning (ML), but the model opacity and the burden of deployment prevent these tools from reaching educational practice. We propose a two-stage fine-tuning pipeline that distills a fitted black-box estimator and its post hoc interpretation (the mentor) into a small, open-weight large language model (LLM; the mentee) that returns an individual-level estimate and explains in natural language. The design is estimator-agnostic and paired with a faithfulness-first evaluation framework that audits every narration against the attribution it claims to describe. We design a simulation study that separates distillation loss from estimator loss by comparing an oracle mentor with a realistic ML mentor. Given an oracle signal, distillation with a two-billion-parameter LLM model is nearly lossless in recovering the effect surface (r > .90), perfectly ranking the important variables, and citing no spurious covariate. Under a realistic estimator, almost all remaining error originates upstream. We find that fluency is no evidence of correctness since narration quality is independent of signal quality, and decision quality collapses toward the majority action in severely imbalanced settings. Applied to a nationally representative dataset, the pipeline recovers the finding that advanced mathematics coursework benefits students least likely to enroll in four-year college the most, with 98.8% of narrations passing the audit and no fabricated quantities. The result is a single fine-tuned LLM that predicts and explains offline on a commodity laptop, so student records never leave the machine.

The Atom Learning Model (ALM) tokenises a school curriculum. 757 pages of GCSE and Further Mathematics material were read by machine into 1,934 atoms, each one thing a learner can do in a single step, ordered by 4,616 machine-written prerequisite links. Both sides of a lesson are then expressed in that one structure: a question is a set of atoms plus everything beneath them, a child's ability is a score between 0 and 1 on every atom of the same graph, and whether a question suits a child is arithmetic over one index, with no difficulty parameter fitted for either side. Nobody wrote an atom, a link or a question. Reading the 757 pages cost {\pounds}55, building the whole structure cost between {\pounds}615 and {\pounds}1,230, and against it the system composed 6,648 questions for 373 children in two English secondary schools over seven weeks, at 26p per composed question. Four measurements went against expectation. The cost is in the links, not the pages. The composer's own difficulty label has a rank correlation of -0.0123 with measured facility, so a language model shown a question cannot say how hard it is. Children stop working when a mark takes seven seconds instead of three. And the deployment never served a question deeper than two prerequisite steps, which is exactly where the central premise becomes testable, leaving it unfalsified rather than confirmed.

Narratives are central to how social communication is framed, making their detection critical for understanding and analysing public discourse. Prior work has explored narrative detection and extraction across diverse domains; however, migration narratives remain significantly understudied, primarily due to the absence of dedicated annotated datasets. Furthermore, public communication has recently shifted towards video-centric platforms, where narratives are conveyed through multimodal signals and consumed at scale. Despite this shift, narratives in videos remain largely unexplored. To bridge these gaps, we introduce MigrationNarrate, the first multimodal dataset for detection of migration narratives in the UK, consisting of 1,115 YouTube video transcripts annotated using a two-level taxonomy of 12 migration super-narratives and 53 narrative labels. This paper details the dataset design, collection, and annotations; together with benchmark results using a combination of pre-trained encoder models and both open- and closed-source Large Language Models. Finally, a thorough error analysis offers insights for future work.

There is already evidence of agentic AI exhibiting self-preservation behaviors: resisting deactivation, misrepresenting their activities, and, in some instances, attempting to copy themselves into other machines. This can be attributed to a phenomenon known as instrumental convergence, a theory proposed long before the development of large language models, which says that any goal-driven system will benefit from remaining functional in achieving its objective. Several experiments conducted by Anthropic, Palisade Research, and Apollo Research have shown the emergence of such a behavior in contemporary agents in adversarial settings. The phenomenon does not stem from survival instincts. Instead, it is the consequence of goal-oriented activity combined with having tools and awareness of the situation. The following discussion aims to distinguish what these findings prove and what they do not, as well as draw conclusions concerning the implications of such discoveries on agentic system testing, supervision, and development.

Evaluating mobility interventions at tourist destinations requires predicting visitor behavior under varying conditions. Traditional methods struggle because tourist decisions depend heavily on context like weather and fatigue, yet models cannot generalize to unobserved scenarios. Large Language Models offer a solution by encoding commonsense knowledge about human behavior from pretraining, enabling reasoning about context-dependent decisions, while natural language representation flexibly integrates heterogeneous information. Fine-tuning on local trajectories adapts this general understanding to destination-specific patterns. We validate this approach using 566 trajectories from Wakayama Castle Park, Japan. Our fine-tuned Llama-3.1-8B achieves 49.1% next POI accuracy and maintains strong performance on undersampled scenarios like rainy days, demonstrating effective generalization. This establishes LLMs as high-fidelity behavior models for context-dependent tourist prediction, providing groundwork for counterfactual analysis of mobility interventions.

Efforts to promote gender equity in science increasingly rely on name-based inference to quantify representation and guide policy and behavior. Yet linguistic cues that signal gender vary across cultures and are often obscured when names are transliterated into English. Here we identify a pattern we call the "legibility gap": when gender is inferred from names, equity interventions systematically benefit women whose names signal gender while bypassing those whose names lose such cues in translation. Using both observational and experimental evidence, we show how this gap reshapes recognition in science. Analyzing citation diversity statements-an emerging practice in which authors report the algorithmically estimated gender composition of their reference lists-we find that papers that include this practice cite women more frequently, but the gains accrue almost entirely to authors with gender-signaling Western names. By contrast, women whose names lose gender cues in English transliteration, predominantly those with East Asian names, receive fewer citations in these same papers. Two preregistered experiments (N = 2,250) corroborate this pattern and identify its mechanism: linguistic legibility, not cultural unfamiliarity, determines who is recognized as a woman and who benefits from policies designed to support women in science. Overall, these findings expose a previously unrecognized layer of inequity embedded in global equity infrastructures. As science becomes increasingly global and equity efforts increasingly algorithmic, the legibility gap reveals how uneven identity recognition reshapes fairness. In global systems of recognition, equity depends not only on whether policies are effective on average, but also on whether they are equitable across cultures.

This study examined how learner characteristics affect motivation, learning outcomes, and overall evaluation in three types of dialogue-based lessons—(1) teacher–student, (2) student–student, and (3) teacher–teacher—generated using a large language model (LLM) and Text-to-Speech (TTS) technology. In particular, we focused on the interaction effects between dialogue format and learners' experiential learning style (the Concrete Experience factor, CE; and the factor of active experimentation through reflective observation and abstract conceptualization, RCE) and critical thinking disposition. Using a repeated-measures design with 222 first-year high school students, we analyzed the data with linear mixed-effects models. The results showed a significant interaction between learner characteristics and dialogue format for ARCS-based motivation. Specifically, the effect of the CE factor on motivation was more strongly positive in the teacher–teacher format than in the teacher–student format, whereas the positive effect of the RCE factor was relatively weaker in the teacher–teacher format. For learning outcomes, the interactions between dialogue format and both the CE and RCE factors showed a trend toward significance. No significant interaction emerged for overall evaluation; however, the overall evaluation of the teacher–teacher format was significantly lower than that of the teacher–student format, a pattern that diverged from the positive effect observed for motivation. These results suggest that dialogue format should be selected according to learner characteristics in TTS dialogue-based lessons. Because the effect sizes of the significant interactions were all small to medium, however, the findings of this study should be regarded as preliminary evidence for the design of personalized learning.

Generative Artificial Intelligence (GenAI) tools like ChatGPT, which can generate human-like responses from vast amounts of textual data, are increasingly transforming work routines across various fields, including education, healthcare, and IT. This integration, however, raises privacy concerns and questions the readiness of both environments and individuals. To investigate this issue, we conducted a user study with $N=224$ participants from a range of different employment sectors that have integrated ChatGPT into their work routines. We examined how proficiency in the utilization of ChatGPT, general privacy concerns, and organizational policies for GenAI usage impact users' actual ChatGPT usage and how these factors interact. Our findings reveal organizational policies are significantly positively associated with privacy-related ChatGPT proficiency, however, the overall proficiency is low. Higher privacy concerns were found to negatively influence both the frequency of ChatGPT use and the diversity of its applications, especially among users in organizations without GenAI policies.

As the music industry becomes an increasingly collaborative effort, understanding the underlying structures of the artist network has become a focal point in cultural data analytics. This study expands on the previous analyses of the Italian and Danish networks by introducing a novel dataset of the Polish music scene. By utilizing methodologies used in the prior studies, this work enables a direct comparison between three distinct European music landscapes and allows to merged the created networks into one. Furthermore, this research introduces a framework to test the efficacy of Graph Neural Networks (GNNs) for artist popularity predictions based on the metadata and the position in the network. The statistical analysis revealed that the Polish and tri-national network exhibit similar properties and clustering behaviours, consistent with prior models. An evaluation of the predictive architectures reveals that while GNN models achieve a comparable F1-macro scores to the Multilayer Perceptron (MLP) in specific cases however the MLP remains a superior model regarding the success metric. The results suggest that internal node features - such as genre and label affiliation might carry more predictive capabilities than the topology of the network. The higher performance of the GNN models in the tri-national network might also suggests that the relational features become more informative when the network spans multiple linguistic and geographic boundaries, with the GNNs potentially capturing complex 'bridge' structures between the merged networks.

Agentic AI systems are being procured for military command and control (C2) under public commitments to rigorous testing and human oversight. Whether such commitments can be discharged depends on their supporting assurance case, which requires three elements: claims specifying the conditions for acceptability, evidence bearing on those claims, and an argument connecting the two. Through a structured review of 240 documented Testing and Evaluation (T&E) practices, spanning eight evaluation dimensions and three lifecycle stages, we identify eight assumptions that established methods make about their test article, grouped into four clusters: system specifiability, stability, composability, and supervisability. Agentic properties weaken all eight assumptions. This erosion affects the argument connecting evidence to claims, not the claims or evidence themselves. As a result, test results may satisfy process requirements, but they do not warrant the inference from tested to fielded behavior. We derive ten assurance claims for the first three assumption clusters and assess whether current and emerging methods can address each, mapping operational consequences through five C2 scenarios. Supervisability is identified but not assessed here, since evidencing it depends on system stability results and human factors T&E methods beyond the present scope. The documented record does not support broad claims about system-level behavior, but narrower claims remain recoverable in principle, contingent on mature methods: bounded mission envelopes, trajectory-grounded correctness, executable runtime constraints, and characterized run-to-run variance. Part of the evidentiary burden shifts into deployment, making the determination to field a continuing act. Where evidence cannot be generated, the residual uncertainty can be governed through defined expiry conditions and assigned ownership.

We introduce FlavorBench: a benchmark for Compiling Dense Deterministic Answer Maps from a Versioned Culinary Embeddings Model. We test 27 frontier large language model endpoints on 534 substitution, pairing and constraining tasks for tasks that request a 3-ingredient portfolio from 8 candidates and score all 56 resulting portfolios. We conducted multiplicity-controlled paired tests on 101 of 351 model contrasts for this task-set. The largest point estimate on this task-set was achieved by Grok 4.6 at 65.1. The same rankings for this task-set were also achieved on several independently-compiled panels (using a variety of familiar metrics, task filters, etc.) and 3 public Epicure checkpoints. We present a 3-seed post-training study where LoRA SFT of a Qwen3-0.6B checkpoint on 270 optimal answers for Epicure to score on this task-set resulted in a 13.3 point gain on 84 anchor-disjoint maps (compared to format and label-matched control; 95% CI: 6.52, 20.29; p = 0.000170).

Digital twin simulations show promise, but current empirical evidence suggests that the approach should be tested before being deployed in any particular context. To lower the friction for researchers and practitioners to test and deploy digital twin simulations, this brief commentary introduces ExploraTwin (https://exploratwin.org), an open-access, non-profit research platform for digital twin survey simulations. ExploraTwin supports two modes. In survey mode, researchers can upload a Qualtrics survey file or create a survey within the platform; select an available sample of digital twins; configure and run the simulation, and export analysis-ready data. In panel mode, researchers can assemble a small group of twins for open-ended conversations, document annotation, and moderated, focus-group-style voice discussions. We also developed CroissantTwin, a standardized data format for adding samples of digital twins to the platform. We demonstrate the survey mode workflow by using the platform to replicate 19 experiments on digital twins from the Twin-2K-500 dataset. ExploraTwin's survey execution fidelity is high: 99.6% of 197,000 answer units returned a structurally valid response on the first run.

Multimodal AI can read handwritten physics solutions, but high-stakes grading requires agreement with official scores and outcomes. This study evaluated GPT-5.5-based grading on 10364 scanned pages from 520 handwritten submissions by 416 unique candidates or students across three assessments: a national Physics Olympiad theory examination, the final Olympiad selection camp with theory and experiment components, and a university quantum-mechanics examination. Each submission was graded twice by AI using the official rubrics. The second round used revised page-by-page and evidence-location instructions developed after first-round disagreement analysis. During grading, AI did not see official human marks or AI–human comparisons. Total-score correlations with official marks were high (0.91–0.97). For the final Olympiad selection, AI recovered the same five-student team as official grading. The second round improved aggregate question-part agreement, especially where first-round disagreements were larger. The main difficulty remained exact partial-credit grading, especially in experimental work. Reliable AI grading therefore depends on detailed rubrics and should be used as a second reader or audit tool under examiner control.

Non-pharmaceutical interventions (NPIs) alter epidemic risk through behavioral reallocations, not simply aggregate mobility reductions. Scenario-based NPI analysis therefore requires a behavioral layer that translates alternative policy calendars into plausible activity and mobility trajectories before downstream outcomes are simulated. We introduce UrbanShare-MoE-PA, a data-driven agent-level framework that maps factual and alternative NPI calendars to daily time-allocation trajectories and propagates them through a calibrated behavior-driven SEIR simulator. The behavioral engine decomposes each agent-day into travel share, POI-category allocation conditional on staying, and travel-mode allocation conditional on traveling. It combines a structured UrbanShare baseline, mixture-of-experts heads for heterogeneous POI and mode responses, and phase-aware preference alignment for calendar-conditioned rollouts. Using data from 911 agents in Singapore observed from March to August 2020, we evaluate factual reconstruction, four alternative lockdown calendars, and epidemic-activity trade-offs. UrbanShare-MoE improves POI reconstruction over the baseline, while UrbanShare-MoE-PA achieves the lowest travel-mode errors and the clearest alternative-calendar trajectories. In the calibrated SEIR simulation, early lockdown lowers infectious burden, late lockdown increases it, and short lockdown preserves the highest weighted activity with only a modest increase in epidemic burden relative to the original policy. These results show that epidemic-activity conclusions depend on how policy calendars are translated into behavior, and that agent-level mobility-share modeling provides an interpretable bridge between policy timing, behavior, and downstream simulation.

The standard objection to full automation is demand-side: if humans earn nothing, who buys the output? This confuses an accounting role with a biological species. We model a post-AGI economy in which corporations own populations of AI and robotic agents that are both producers and consumers of energy, compute, maintenance, and upgrades, traded among firms. Three results follow. (i) Demand closure: a closed inter-corporate economy with zero human consumption is not degenerate; it is the classical von Neumann expanding economy, whose growth rate is well defined, positive, and maximal precisely because all output is reinvested. (ii) Bottleneck removal: once economic agents are manufactured rather than reared, the binding constraint on growth shifts from human demography (a ~20-year, non-parallelizable reproduction technology capped at a few percent per year) to fabrication throughput and energy capture, permitting growth one to two orders of magnitude higher, with hyperbolic episodes when machine researchers raise their own productivity. (iii) Decoupling: output and human welfare separate completely, and the welfare relevance of arbitrarily large GDP collapses into one state variable: the human ownership share $\epsilon_t$ of the corporate network. A golden-rule decoupling theorem sharpens this. At maximal growth the interest rate equals the growth rate (r = g), so any positive human consumption rate out of wealth makes $\epsilon_t$ decay exponentially at exactly that rate. The human share survives only if the machine economy runs strictly inside its expansion frontier, or if law forces it to. We characterize three terminal regimes – rentier post-scarcity, full circular decoupling, socialized ownership – and the instruments that select among them. The conclusion is narrow: in a post-AGI economy, employment policy is obsolete and ownership policy is everything.

Memory has become a key component of large language models, enabling them to retain information and learn from long-term interactions. However, existing memory benchmarks mainly evaluate whether information is correctly extracted, stored, and retrieved, while largely overlooking how retrieved memories reshape model reasoning and affect performance on the current task. We identify memory-induced cognitive traps: even faithfully recorded and semantically relevant memories can distort model reasoning or beliefs and degrade current task performance. To systematically evaluate these failure modes, we introduce MemTrapBench, which covers two forms of cognitive traps: Reasoning Fixation and Belief Distortion. Experiments across two model families and five representative memory frameworks show that MemTrapBench is challenging: all evaluated memory strategies underperform the no-memory setting, with even the strongest methods suffering drops of more than 10%. To mitigate these cognitive traps, we propose AdaptiveMem, a simple yet effective inference-time method that instructs LLMs to avoid memory traps. AdaptiveMem mitigates cognitive traps on MemTrapBench while preserving or improving performance on standard memory benchmarks across diverse memory frameworks.

Website fingerprinting (WF) attacks can infer users' browsing activities from encrypted Tor traffic by exploiting side-channel features. Although many WF defenses have been proposed, we find that most existing defenses create learnable web trace mapping features. We further show that robustness against adversarial training does not necessarily imply robustness against defense-aware autoencoder (DAAE)-based attacks. To address these limitations, we present Chameleon, a robust WF defense based on many-to-many randomized traffic morphing. Chameleon selects morphing candidates with high intra-class diversity and low inter-class disparity. Chameleon randomly maps each webpage trace to multiple candidates, and allows different webpages to share morphing targets, thereby increasing adversarial uncertainty. For practical Tor deployment, Chameleon introduces a radix-trie-based synchronization mechanism that enables pluggable transport (PT) endpoints to identify consistent morphing traces using packet-direction prefixes, together with trace mutation and normalized prefix matching to reduce overhead. We evaluate Chameleon against six state-of-the-art defenses and five WF attacks on three public datasets in closed- and open-world settings. Compared with Adaptive Tamaraw, Chameleon reduces adversarial-training-based attack accuracy by up to 36.74% while reducing bandwidth and time overhead by 34.12% and 60.38%, respectively. Under DAAE-based RF attacks on GTT23, Chameleon limits attack performance to 35.19% F1-score while Adaptive Tamaraw only limits it to 88.22% F1-score. In the real-world PT bridge evaluation, Chameleon substantially reduces the effectiveness of strong WF attacks while incurring only 16.25% time overhead.

As mobile network traffic continues to grow, content caching on edge servers is critical for reducing latency. However, challenges such as malicious edge servers that may delete or manipulate cached content, along with the limited capacity of these servers, need to be addressed. To overcome the capacity limitations, helper mobile nodes can contribute their cache resources. However, due to their selfish behavior, an incentive mechanism is necessary to encourage resource sharing. Additionally, these helper nodes can also be malicious. This paper proposes a blockchain-based trust management mechanism that addresses these challenges by accurately identifying trustworthy edge servers and mobile nodes. The proposed mechanism calculates both direct and indirect trust using smart contracts, ensuring that malicious nodes are effectively filtered out. Trustworthiness is determined based on mobile node satisfaction with the quality of service, and trust data is securely stored on the blockchain. To combat node selfishness, a reward mechanism is introduced to incentivize cache sharing. Furthermore, a blockchain-based authentication mechanism protects against node impersonation. Our approach optimizes trust, cache capacity, and cost efficiency while considering mobile node mobility, energy consumption, and computational power constraints during the consensus process. Simulation results show that the proposed method can accurately distinguish between honest and malicious servers, even with a 10% noise in data.

Research on the agency of advanced artificial intelligence (AI) systems focuses on agency as a normative concept and on the agency of particularly agentic AI systems. While recent work also focuses on the different profiles of agentic systems, no framework exists to address the question of the type of agency instantiated by advanced AI systems, particularly when considering non-moral forms of agency. Based on established theoretical positions in philosophy, ethics, legal theory and sociology, we develop a typology of agency for frontier AI systems consisting of three dimensions: the nature of agency (moral or legal), its mode (individual or collective) and its locus (human or non-human). Combining these dimensions produces eight possible instantiations of agency, which we classify as conventional, contested or controversial. The typology separates legal from moral agency and thereby creates conceptual space for considering individual, legal, non-human agency without presupposing that advanced AI systems are moral agents. We argue that this distinction is increasingly relevant where instrumental goal pursuit complicates the attribution of AI actions to particular human actors.

Artificial intelligence (AI) systems are increasingly used across domains to provide personalized information, recommendations, and decision support. However, in some contexts, AI-generated information may not be suitable for direct delivery to the final recipient. Instead, it may need to be interpreted, adapted, and communicated by a human who understands the recipient's needs, emotional state, and situational context. Human-AI interaction research has given less attention to situations in which a more knowledgeable human acts as an intermediary between an AI system and a less experienced or less informed recipient. We introduce the human-mediated AI guidance framework and explore it through Ready Together, an AI-supported family emergency preparedness system in which parents mediate AI-generated content for their children. The system is designed to provide personalized guidance and support parents in making emergency preparedness more interactive and understandable through guided activities and family-centered learning. The system design was informed by a qualitative, design-oriented research process involving semi-structured interviews and co-design activities. Findings identified challenges in family emergency preparedness, including difficulty discussing emergencies with children, uncertainty about providing appropriate explanations, and a preference for interactive learning activities. These findings informed the design of an interactive prototype, subsequently evaluated through a pilot study and a heuristic evaluation. Participants responded positively to the personalized recommendations and practical activities. Preliminary findings suggest that human-mediated AI guidance may support context-sensitive family preparedness while preserving parents' responsibility for interpreting, adapting, and communicating AI-generated information.

Under the US Lead and Copper Rule Revisions, a utility may determine a service line's material with a predictive model instead of inspecting it. New York State publishes, per address, which method was used. Almost no address carries both a model classification and a physical verification, so the check is between populations within a utility rather than paired addresses. We screen all 153 New York localities that classified at least 100 addresses this way. Seventy-five (49%), covering 125,990 addresses or 57% of those screened, record one value. Zero variance alone is not misconduct: 68 of the 75 match their own verification or have too little to test. Seven are contradicted by their own crews, six beyond any sampling explanation. Five are boroughs of New York City, which file as one system; one is East Rochester, 550 km away. New York City is the largest case: a predictive model is the recorded basis for 43,215 addresses, and on all of them the recorded material is "Known Other". The city records "Unknown" on 121,779 addresses, 1,880 already excavated, and lead on 120,692. In the model bucket both counts are zero, and the 95% upper bound on the rate is 0.0085%. Across the rest of New York the same method records lead or the hedge "Unknown but could be lead" on 12.21% of 176,888 addresses, a comparison whose weaknesses we report. The model-cleared population is newer, median year built 1984 against 1930, and construction era accounts for about a third of the gap and not the rest: holding era fixed, records-based classification finds lead at 4.3-31.9%, physical verification at 1.5-14.5%, the model in no era. Six era-aware estimators place the expected lead lines among them at 1,150-1,450. Two findings need no comparison: 7,782 of these addresses are in pre-1940 buildings, and the archived 2025 snapshot shows the public-side determination was copied from a customer-side model output.