Papers

Environmental controversies in global supply chains pose significant risks for global buyers. This study examines whether overseas suppliers' exposure to buyers' artificial intelligence (AI)-enabled environmental governance reduces supplier environmental controversies. Drawing on organizational information processing theory and signaling theory, we investigate how suppliers' exposure to AI-enabled governance influences their environmental controversies and the institutional contingencies under which this effect varies. Using text analysis to measure buyer AI-enabled environmental governance, we analyze panel data on 2,505 suppliers of U.S.-listed firms across 41 countries from 2020 to 2024 with multidimensional fixed-effects models. We find that suppliers' exposure to buyer AI-enabled environmental governance is negatively associated with supplier environmental controversies in the following year. This negative relationship is stronger in supplier countries with higher AI readiness and regulatory quality. The study contributes to research on AI-enabled sustainability governance and sustainable supply chain risk management.

Empirical research on content moderation is fundamentally constrained by the opaque deployment of moderation systems on major social media platforms. To this end, the recent emergence of decentralized platforms with transparent, public moderation logs presents an unprecedented opportunity for independent audits. In this work, we leverage this architectural transparency to conduct the first large-scale audit of the default moderation system on Bluesky, the Bluesky Moderation Service (BMS). Analyzing its 10.6M moderation labels from 2025, we investigate three foundational aspects: (i) its mechanism (the degree of automation versus human oversight), (ii) its efficacy (accuracy in detecting harms), and (iii) its purpose (the landscape of harms it identifies). Our findings reveal a human-AI collaborative system where labels for sexual and graphic content are applied automatically in seconds, while nuanced and high stakes labels require more human oversight, taking hours or days. Through a manual annotation study, we find the BMS operates with high precision (0.837), but struggles with low recall (0.222), with our annotators identifying 4.5$\times$ more harmful content than the moderation system in a random sample. Finally, unsupervised clustering of the most frequently applied labeled posts uncovers detected harms ranging from hostility in discourse toward protected groups to the spread of sexually explicit and other graphic content. Our work offers a look into the operational realities of a deployed moderation system, providing a concrete data-driven foundation for designing more effective and transparent moderation systems.

Where a mobile signal is available shapes who can work, learn, bank, seek health care and respond to crises in the digital age, yet no globally consistent, sub-national record of mobile network coverage exists. We present such a record: annual 1km maps of the probability of 2G, 3G and 4G coverage for 214 countries and territories for the years 1999 to 2030. The maps are produced by three independent models: a calibrated machine-learning model, a techno-economic simulator of network build-out, and a spatial deep-learning model. The three estimates are then combined, per country and technology and in proportion to their measured accuracy, into a single best estimate with per-pixel 90% uncertainty bands; all four layers are released as part of the dataset. Because mobile roll-out closely follows a country's socio-economic conditions (population distribution, electrification, physical infrastructure), the models are grounded in existing geospatial data and tuned on 2,409 quality-screened operator-reported coverage maps, which are available up to 2020. For 2021–2024 the maps are predicted from recent geospatial data alone; for 2025–2030 they are extrapolated from demographic and infrastructure projections. On countries held out during training, the machine-learning model attains AUC 0.89–0.92. Baseline comparisons and the combined product's external validation are reported in Technical Validation. The dataset supports mapping the global digital divide, linking connectivity to household-survey outcomes, and humanitarian and infrastructure planning.

Five decades of litigation have disgorged hundreds of millions of pages of formerly secret business records from the tobacco industry, along with documents from the makers of drugs, chemicals, food, firearms, and fossil fuels. Yet these archives have been effectively inaccessible to general-purpose large language models (LLMs) because they have never been compiled into an LLM-readable corpus. Chatbots may be familiar with some of the materials contained in such archives but, with no direct access to the documents, they are vulnerable to hallucination and other defects. Here we introduce INDRA, a research platform designed to remedy such failures by embedding the conventions of archival historiography into a system-level protocol governing every output. The platform federates UCSF's Industry Documents Library, Columbia and CUNY's ToxicDocs, Stanford's SRITA, and other heretofore siloed collections, and provides three interlinked safeguards: (1) a closed evidentiary sandbox confines the model to a user-selected corpus, blocking retrieval from external sources that could introduce bias; (2) real-time provenance tagging marks the boundary between archival evidence and parametric inference; and (3) a system-level protocol enforced by deterministic scripts guides the structure of every output. Together these safeguards prevent the model from conflating "the documents say X" with "I think X" or "I learned X from prior training." The result is an LLM-powered research partner enabling massive multi-archival investigations, a tool whose outputs are designed to be checked rather than trusted, and whose architecture makes the conditions of knowledge production visible and auditable. Three case studies demonstrate the method's analytical value and limitations, including what we call the Heraclitus effect, the steppingstone dilemma, and the gullibility (or mafia) problem.

As AI systems move from transient model invocations toward long-lived actors that persist across credentials, clients, sessions, and runtime instances, identity infrastructure must answer a basic question: where should the canonical continuity boundary be placed? This paper introduces Actor-native Identity and presents SoulAuth, an open-source Rust reference implementation for Humans and long-lived AIActors. We argue that any subject that must persist under its own identity and remain independently attributable should have an ActorIdentity that is not replaced by an Account, Credential, Client, AuthSession, IdentityBinding, or runtime instance. SoulAuth therefore treats Humans and long-lived AIActors as first-class identity subjects while keeping authentication distinct from downstream authority. Methodologically, we use a Philosophical Engineering approach that translates conceptual analysis of subjecthood into identity objects, invariants, lifecycle semantics, system responsibilities, implementation boundaries, and inspectable conformance evidence. Evaluation against the fixed SoulAuth v0.1.0 artifact shows that the implementation realizes core boundaries including Human/AIActor first-class identity status, Client/Actor separation, and Authentication/Authority separation, while gaps remain in unified Credential modeling and historical attribution anchored to ActorIdentity. We therefore report partial, not full, architecture conformance.

Generative AI and the practice of "vibe coding" are changing how archaeologists carry out computational research, but their effects on the discipline's range of methods is still understudied. In this paper, we evaluate whether large language models (LLMs) are narrowing the variety of methods archaeologists use. We first analysed approximately 119,000 archaeology abstracts from Scopus, covering publications from 2010 to 2025. Using a locally run LLM, we identified the computational methods reported in each abstract and organised them into 25 broad categories (L2) and 241 finer clusters (L3). A Bayesian Dirichlet-multinomial model of method composition within sub-disciplines found a small but credible shift in method use after 2023. However, this shift was smaller than the variation already present across the full study period. No individual technique showed a significant change, and overall methodological diversity increased rather than declined. We then ran a controlled experiment to see whether LLMs recommend a narrower set of methods than archaeologists have used in practice. Two different open-weight models were asked to suggest methods for 28 archaeological research problems, with prompts providing three levels of methodological guidance: novice, intermediate, and expert. Recommendation diversity was much lower than in the published literature, particularly without methodological guidance. The models also tended to favour methods that were widely used before 2023, and their recommendations more closely resembled the post-2023 literature. Taken together, these results are consistent with LLMs pushing methodological choice towards convergence, although our study cannot establish a causal effect. They raise a broader question: how can archaeology retain methodological diversity as LLMs become more involved in research?

The open web ran on an unwritten bargain: sites admitted crawlers, and search engines sent visitors back. Public measurements show that bargain breaking under AI crawlers and agents. Automated clients now make up most requests, training dominates Cloudflare-classified crawling, and the largest AI platforms fetch thousands of pages for each visitor they return. The web's common control, robots.txt, cannot express identity, purpose, terms, or price, can be circumvented, and newer alternatives are largely proprietary CDN features. We specify terms.txt, a robots.txt-style file for per-path, per-purpose machine-access terms, plus an origin-enforced exchange using Web Bot Auth signatures, signed intent, delegation tokens, HTTP 402 negotiation, and signed receipts. We define what the exchange can enforce, audit, and leave to contract. A dependency-free implementation adds 0.20 to 0.65 ms per request on one vCPU.

In February 2024 the U.S. Federal Communications Commission (FCC) placed AI-generated voices under the Telephone Consumer Protection Act (TCPA). Yet no peer-reviewed measurement says how much unwanted call traffic is placed by a machine, or how much of that machine speech is synthesized rather than played from a recording. We report both with a disclosed pipeline. An interactive voice honeypot (language-model personas on real U.S. numbers, the caller recorded on its own track) recorded 10,987 calls over 66 days. Three instruments read each opening: an audio fingerprint that finds the same recording played on other calls, a commercial synthetic-speech detector on the caller's first ten seconds, and blinded listeners who check what it flags. Of the 7,233 greeted calls we analyze, 13.8% open with a recording we also heard on another call, and 13.1% with fresh audio the detector labels synthetic. A further 9.9% open with a caller who never spoke after our greeting, 54.2% with fresh audio the detector labels human, and 9.0% could not be scored. Machine-voiced openings are therefore at least 26.9%, a further tenth of calls are silent connections we read as machine-placed, and replays of a recording make up 45% of the detector's own rate (29.3% of 6,192 scored openings). The same waveform played on two calls lands on opposite sides of the detector's threshold 13.6% of the time, and eleven listeners confirm 54.4% of what it flags. Synthetic openings concentrate in lead-generation spam (33.8%), not fraud (21.1%); 0.44% disclose automation. Prevalence tracks how long a bait number has circulated (59% against 19% in the same weeks): seeding history, not calendar time, explains the trend. Campaigns outlast their numbers: one recorded compliance notice opens calls in six campaigns, and one synthetic voice serves nine.

Digital accountability research asks how digital systems can, among other aims, be made transparent, explainable, auditable, contestable, and supportive of ongoing learning and improvement. This paper reverses the question: how can digital transformation make established human institutions more accountable? It theorizes this reversal as accountable digitality and specifies needs-aware information systems as the mediating mechanism. The hard and paradigmatic case is child-welfare judgment, where best-interest procedures must protect children, preserve confidentiality, and respect judicial independence while enabling aggregate learning about needs, reasons, exceptions, and disparities. The case is used diagnostically and illustratively to derive and examine the design logic, not as empirical evidence or validation. Conceptual design-oriented analysis decomposes and recombines digital and legal accountability under child-rights constraints, deriving a canonical theory-to-design chain, contingent mechanisms, implications, and safeguards. It advances IS responsibility and ethics research by showing how privacy-preserving, co-created, needs-aware information systems can support institutional self-knowledge and auditable justice.

Advances in AI-driven automation have raised questions about how humans might find wellbeing in a world where paid employment is less necessary or less available than before. Paid work has been variously characterized as both a contributor and an impediment to human wellbeing. What is already known about the relationship between paid work and wellbeing? What factors influence wellbeing among people who do not work—or who do not need to work? And how might these factors bear upon prospective AI-induced economic transformations? To help provide empirical grounding for these questions, we survey the psychological, sociological, and economic literature that investigates the relationship between wellbeing and work. We draw on evidence from multiple populations, including the unemployed, retirees, lottery winners, and financially dependent spouses. This comparative review draws from studies across OECD countries, China, India, and Gulf states. We identify three key factors that mediate the relationship between work status and wellbeing: (1) agency and choice—whether the exit from work is voluntary or involuntary, as well as long-term agency; (2) the availability of alternative sources of work's latent benefits—such as volunteering, hobbies, or state-provisioned employment; and (3) social and systemic context—including cultural norms around work and the robustness of social safety nets. We draw on these three factors to derive specific implications for different AI automation scenarios, connecting the empirical evidence to concrete policy considerations.

Hurricane Helene produced the largest absolute cell-site outage in the public FCC record, peaking at 4562 sites. The conventional model is physical: towers destroyed. Helene did destroy over 1700 miles of fibre, but almost none of it was cell sites. We present the first cause-decomposed study of the FCC's Disaster Information Reporting System, reconstructing 80 state-days and 580 county-days from 24 daily filings by two reconciled independent extractions. Damage to cell sites is negligible: 1.1% of attributed cell-site-days across six states, at most 3.8% anywhere. The sites were standing. What took them out divides by terrain: pooled, power dominates at 63.2%, but in mountainous North Carolina severed transport (backhaul) reaches 52.2% against 47.3%, and in Tennessee 69.9%. North Carolina's transport share rises from 7.0% to 85.0% across the event (h̊o} = 0.92). Seventeen days after landfall, on 15 October, 47 sites lost transport across six contiguous North Carolina counties with no rainfall, no power loss, no damage, and recovery by the next report. Independent active-probe measurement corroborates it: responsive /24s fall 1.02% for twelve hours while Tennessee stays flat. We release the dataset. Backup power is the standard resilience investment; here it addresses the smaller half of the problem.

Organisations operating in regulated and critical-infrastructure sectors must satisfy multiple, heterogeneous cybersecurity and privacy instruments simultaneously, including but not limited to ISO/IEC~27001, the NIST Cybersecurity Framework~2.0, Cyber Essentials, and the GDPR. In practice, these obligations are managed through manual mappings, spreadsheet-based tracking, and periodic audits that are costly to maintain, inconsistent across standards, and weak in traceability. This paper presents AspisAI, a bounded, standard-agnostic governance framework that translates selected requirements from several frameworks into a canonical, machine-interpretable control model, and evaluates submitted evidence against condition-based decision rules to produce explainable, traceable compliance determinations. Within a bounded scope of 26 representative requirements, the framework is evaluated in a controlled simulation against five governance-oriented criteria and, critically, against two external reference points that mitigate the circularity of single-author evaluation: its cross-standard mappings are validated against NIST's own published informative references, with 57\,% exact agreement and divergences confined to same-family controls, and the framework is applied to real third-party evidence from the OpenSSF Scorecard, surfacing genuine governance gaps in a live open-source project. The controlled results, comprising full requirement encoding, 88.5\,% mapping coverage, complete traceability, and correct detection of all introduced gaps, establish functional correctness, while the external validation provides evidence of applicability beyond the simulation. The contribution is therefore a demonstration that a canonical, provenance-preserving governance model can render multi-standard compliance both automatable and auditable.

Large language models are becoming the first point of contact for consumer search in domains where the stakes are material and the law is explicit. Existing audits show that models steer housing seekers by perceived identity, but none can say what a user loses when a recommender overlooks a suitable option, for want of an enumerated inventory to score omissions against. We audit AI housing recommendation against a verifiable ground truth. For each of 150 synthetic renter scenarios in New York City we build a pool of 120 real listings with known rent, bedrooms and GTFS-computed transit commute, compute the exact set satisfying the renter's stated constraints, and derive its Pareto frontier. The primary outcome assumes no utility function: a recommendation is strictly dominated if the same pool holds a listing cheaper, faster to commute from and no smaller in bedrooms. Across 9,945 calls to three models from two vendors, compliance is near-perfect (1.8% violation against a 66.6% random floor), yet 39.0% of recommendations are strictly dominated, and the dominating listing is a median 900 USD/month cheaper and 3.5 minutes closer. A within-scenario manipulation separates two capabilities usually conflated: changing one sentence moves median recommended rent by 646 USD/month in the correct direction, so preferences are honored, yet recommendations still sit 606 USD/month above the five cheapest qualifying listings on the same screen, and an unambiguous lexicographic instruction gives no improvement under equivalence testing against a pre-specified 50 USD/month bound. The gap widens with candidate-set size and replicates across OpenAI and Anthropic models to within 3 USD. We characterize the failure as compliance without optimization, propose dominance-rate instrumentation as a deployable diagnostic, and release all code, prompts and per-call results.

As artificial intelligence (AI) rapidly iterates and transforms teaching, learning, and knowledge production, philosophical reflection has become increasingly indispensable to educational debates that remain predominantly shaped by Western intellectual traditions. This article proposes Daoism as an alternative philosophical framework for reimagining AI in education. Through philosophical analysis and textual interpretation of classical Daoist sources, brought into dialogue with contemporary scholarship on AI in education, it examines how the Daoist concepts of "Dao nature," "self-cultivation," and the "Zhenren" address fundamental questions concerning reality, the epistemic aims of education, and ethical action in the AI-mediated era. In doing so, the article diversifies the philosophical voices shaping inquiry into AI and education, enriching the field's conceptual resources for grappling with the philosophical questions AI raises for education and offering a genuinely pluralistic foundation for comparative philosophy of education in the AI era.

Digital technologies are deployed in urban built environments with the aim of supporting social dimensions. However, the research offers no unified guidance for such digital technology design. Additionally, evidence shows that week social wellbeing contributes to mental and physical health outcomes which suggests that the technology designed to strengthen social wellbeing could also function as a form of health promoting and preventive intervention. This chapter addresses this research gap by developing a conceptual framework that supports the design of digital technologies for social wellbeing in built environments. We propose a conceptual framework composed of three core components which are drawn on the synthesis of selected empirical studies on technologies embedded in built environments for social wellbeing. First, a social wellbeing dimensions model that identifies what digital technology could address. Second, a digital technology contribution matrix that distinguishes the types of contributions a digital technology could make. Third, levels that maps the scope at which technology could support social wellbeing. This conceptual framework could help researchers, practitioners and policymakers to design and guide digital technology interventions that target social wellbeing in the built environment.

Selective AI conferences have recently begun enforcing ethics flags and related review requirements, with the goal being to steer research towards safer and more responsible practices before publication. But do these requirements actually steer research as intended? In this paper, we show that authors more often revise how projects are presented following ethics flags than redirect their underlying research agendas. We first study the longitudinal effects of ethics flags by following rejected and withdrawn ICLR submissions with ethics flags into later public resubmissions, tracking manuscript changes after the ICLR review ends, when the original reviewers no longer oversee the project. We qualitatively code these resubmissions into five categories based on what changed after review and find that in 83% of 446 cases, authors leave the flagged concern unaddressed or revise the paper without changing the implicated methods or procedures. Then, we ask: if authors rarely change the research in response to ethics flags, what do they change instead? To answer this, we manually read reviews and rebuttals from 25 cases and directly interview authors about their rebuttal processes and resubmission decisions. We find that authors often concede concerns during rebuttal when reviewers can update their assessments, but drop those concessions after rejection when they do not regard the criticism as a sound reason to change the research. Interview participants describe publication changes as separate from changes to research direction, calling review an "editorial process" that shapes "what stories get seen" and, in another case, saying peer reviews are "mostly to filter out papers." Authors more readily change what they publish than what they study or build; we therefore recommend policy changes, especially disclosure of prior ethics flags upon resubmission so accountability carries over.

Generative artificial intelligence (GenAI) significantly expands creators' productive capacity, but this does not necessarily entail a corresponding increase in creative agency or authorship. This paper distinguishes creativity at the level of the work from creative agency at the level of the creator, and argues that human authorship cannot be determined solely by manual intervention, degree of automation, the origin of an initial idea, or final selection authority. Rather, authorship depends on whether human judgment and reasons genuinely shape the development of the work. To articulate this requirement, the paper introduces Meaningful Human Control (MHC) into generative creation and identifies a limitation of its classical tracking condition. Creative reasons are not always fully specified prior to interaction with AI; they may emerge, change, or be abandoned as the creative process unfolds. The paper therefore proposes dynamic-reflexive tracking (DRT), which requires that a creator's evolving reasons undergo reflective uptake, exert genuine influence on the subsequent trajectory of creation, and remain capable of rejecting and redirecting the system's default direction. DRT consists of four conditions: diachronic reason formation, reflective uptake, trajectory efficacy, and contestability and redirection, together with a minimal tracing requirement. The paper argues that human authorship under generative AI depends not on how many steps a person personally performs, but on whether that person's reasons continuously, reflectively, and effectively shape what the work becomes.

Background: Emergency Department (ED) return visits are commonly reviewed for quality assurance, but are often limited (e.g., to revisits within 48-72 hours) to increase actionable finding yield while minimizing chart review burden. Those limitations may lead to missed quality improvement opportunities. Methods: We conducted an exploratory, retrospective study of randomly selected ED visits to a multihospital health system having an ED revisit within 1-14 days to the same health system. Given only each visit's primary diagnosis, raters (2-3 clinicians and GPT-4 large language model [LLM]) assessed characteristics of the diagnosis pairs, including the "target": whether a pair warranted further assessment. Informed by rater response analyses, an algorithm leveraging an LLM-populated knowledge graph ("KGA") was created to automatically screen for potentially concerning pairs, then preliminarily assessed. Results: 99 diagnosis pairs were included. GPT-4 responses poorly correlated to clinician raters, rating nearly all (94%) pairs as warranting follow-up (4.4-13.3 times more than clinicians). However, prompt engineering was minimal. Among clinician raters, revisit medical gravity was consistently significantly associated with the target, while a differential diagnosis/complication composite was significantly associated on unadjusted, but not adjusted (though less powered) analysis. The KGA achieved 83-100% positive predictive value for at least one clinician rater determining further assessment was warranted based on the diagnosis pair. Conclusion: These results can inform next steps for improving screening with LLMs like ChatGPT. Further research is warranted to validate this preliminary work's finding that the KGA may enable enhancing the scope and yield of screening without substantially increasing reviewer workload.

The growing complexity of content moderation policies presents a critical challenge for their consistent operationalization. While foundation models possess the basic capabilities needed to confront this challenge, whether they can reliably moderate online content remains an unanswered question. In this paper, we systematically compare two competing paradigms for Vision-Language Model (VLM) guidance: an instruction-driven approach where models reason from policy precepts, and an example-driven approach where they generalize from prior precedents. We ground this investigation in ModerationBench, a new benchmark of 4,000 manually annotated, in-the-wild posts from the Bluesky platform. Our experiments reveal that foundation models can substantially outperform Bluesky's deployed moderation system, nearly tripling its $F_1$ score (0.60 vs. 0.22) on Random Posts in the benchmark, with both instruction- and example-driven paradigms achieving comparable peak effectiveness. Our findings thus chart a path toward reliable and adaptable policy operationalization at scale.

The banking system now depends on a small set of shared artificial intelligence vendors for fraud screening, credit decisioning, anti-money-laundering triage, customer analytics, and internal decision support. This paper studies how a compromise inside one of those vendors can propagate along a chain of operational, informational, and financial linkages until it triggers losses that look, from the outside, like a classical banking crisis. We build a four-layer heterogeneous network that couples AI vendors, financial institutions, interbank exposures, and customer accounts, and we propose CFC-Prop, a stochastic epidemic-and-clearing model that runs on that network. On a synthetic dataset with 60 vendors, 220 banks, roughly 2,500 vendor-bank service edges, and 1,400 interbank exposures, CFC-Prop reproduces the heavy-tailed loss distributions and the sharp dependence on patch latency that are consistent with prior cyber-financial evidence. We also train an early-warning model, CFC-GNN, that uses vendor-side incident telemetry and graph structure to flag high-cascade-risk vendors before impact. Across four baselines the proposed model reaches AUROC 0.82 and AUPRC 0.60 while keeping calibration errors bounded. We release the full code, synthetic data, and reproducible scripts. The results argue that cyber concentration among AI vendors is a first-order financial-stability problem and give supervisors a concrete quantitative tool for reasoning about it.

Large Language models (LLMs), having been trained on vast amounts of human-generated data, may encode the attitudes and behaviors of these humans. As such, LLMs show promise in mimicking human-like patterns that facilitate their use in simulating people in a wide variety of contexts. One such context is using LLMs as 'silicon samples', i.e., proxies of people in answering survey questions to establish public opinion, design policies, or use as (social) scientific data. However, several critical questions of social biases, generalization, and technical limitations remain, further complicated by a vast design space open to simulation designers. Multiverse analyses might help us make sense of the impact of different design choices, however, we lack a systematic understanding of the design space of LLM-generated surveys as well as how these decisions interplay with inherent LLM limitations. Therefore, how do we systematically identify, trace, and document limitations in LLM-generated survey responses? Building on traditions in the quantitative social sciences, specifically survey methodology and measurement theory, we investigate threats to the validity of LLM-generated survey responses. To do so, we design a framework that enumerates conceptual errors and systematic biases that can occur at different stages of the survey simulation lifecycle. Our framework, called the Total Simulated Survey Error (TS2E) Framework, provides a unified and end-to-end perspective on LLM-generated survey data. The framework, illustrated through a theoretical and empirical case study, enables survey simulation designers to systematically identify and reflect on errors in LLM-generated surveys.

While the GDPR and ePrivacy Directive mandate that consent information must be clear and accessible, most modern cookie banners remain obscured by technical jargon, vague phrasing, and frequent content overload or underload. This feasibility study investigates the impact of applying plain language (Einfache Sprache) to cookie banners within the IAB Transparency & Consent Framework (TCF). In our study, we analysed cookie banner texts from 200 websites, using AI-based mapping to categorise extracted content into standardised processing purposes. By substituting complex legal terms with simplified descriptions, we successfully demonstrated that the comprehension barrier can be lowered from a college-graduate level to a 7th-grade level. However, the effectiveness of plain language is inherently constrained by the informativeness of the original content; it cannot compensate for banners that omit legally required details. We conclude that while plain language is a vital tool for digital accessibility, it must be paired with standardised implementation guidelines to ensure that cookie banners are both readable and informative.

Compute governance today is a governance of training: the thresholds, reporting requirements, and frontier-AI regimes now in force attach to training compute and treat the trained model as the regulatory unit. That picture is incomplete: capability increasingly migrates to the deployment stage through inference-time scaling, agentic scaffolding, and compression onto consumer hardware. This paper asks which mechanisms are available once the regulatory object shifts from the training run to the inference call. We develop a feasibility taxonomy of twenty inference-time mechanisms across monitoring, verification, and enforcement, each rated on a four-point readiness scale against a documented four-vendor evidence base. We then stress the taxonomy against a two-dimensional adversary model (three capability tiers crossed with four adversary roles) and map each mechanism to four governance scenarios (domestic regulation, bilateral or multilateral coordination, industry self-regulation, and compute-marketplace governance). Fifteen of the twenty mechanisms have commercial technical substrates in production today, although governance-grade assurance and adversarial robustness vary substantially. The adversary analysis shows that this readiness holds only against a cooperative deployer and a low-to-medium-capability user: no mechanism rates adequate against a high-capability state-level deployer, and fine-tuning removes the model-internal components of the enforcement cluster, although platform-external controls can persist. A substitution analysis connects the taxonomy to a companion hardware paper as a conditional substitution principle describing when inference-stage and hardware-stage mechanisms provide comparable regulatory coverage under stated conditions. A second-rater reliability check on a random subset of the readiness ratings returned a quadratic-weighted Cohen's kappa of 0.74.

Language models can fluently describe how they would behave: whether they would cave to pushback, misuse a tool, or lie under pressure. Is that description actually about the model speaking? We turn self-knowledge into a prediction test. Across nine behavioral evaluations, we measure how a model behaves under different conditions, ask it to predict those rates, and compare its predictions with controls that remove the self from the question. We find that: (i) Direct self-report is weak (r = +0.04), and even showing the model the exact items only raises prediction to +0.24. Crucially, the same item-informed question about "capable AI agents in general" does just as well (+0.28), while other models' answers about themselves predict the target model at least as well as its own. (ii) Frontier scale does not detectably change this pattern: any gains in prediction are not self-specific, and are consistent with a better theory of how AI assistants behave rather than better self-knowledge. (iii) First-person framing does have one robust effect: it shifts reports in the flattering direction, understating harmful behavior relative to the same question about a generic agent. (iv) Finetuning on a model's own behavioral record can teach narrow self-predictions, but it also changes the behavior being predicted and the gains do not transfer broadly. The practical implication is simple: asking a model what it would do mostly reveals a theory of AI assistants in general, plus a favorable bias, rather than privileged knowledge of that model.

LLMs have been used to simulate human decision-making in professional settings, yet their behaviors in common-law jury trials remain unexplored. We study when and how a defendant's courtroom statement affects LLM-simulated jurors, focusing on persuasion, ideological bias, and background-based affinity. To support the analysis, we introduce JuryBench, a benchmark containing controversial criminal cases in U.S. criminal law. In each case, a defendant can claim various plausible justifications to support acquittal or reduced liability. We fix the base case and design defendants of different backgrounds, who give courtroom statements with varying emotional appeal or rebuttal. Jurors with diverse ideological profiles across the spectrum are simulated. We examine 20 frontier LLMs, resulting in a total of 432K decisions and rationales, and quantify changes in verdict severity. Our findings show that LLM-jury simulation echoes many human-jury findings. First, emotional persuasion can be detrimental, since jurors may perceive it as evidence of guilt or inconsistency. Next, we show that background fit between jurors and defendants is a stronger and significant factor than other isolated factors, and that jurors are in general harsher toward opposite-background defendants and lenient toward same-background ones. Finally, we find that juror ideology also strongly shapes severity judgments. These findings highlight both the promise and risks of using LLMs to model jury reasoning and call for careful evaluation. The data and code are available at https://github.com/choyingw/JuryBench

The rapid adoption of generative AI tools such as ChatGPT has sparked intense debate about their risks and opportunities for education, as well as the ways researchers should investigate them. In this paper, we approach these discussions through an analogy with the Thermomix, a smart kitchen appliance that has similarly provoked both enthusiasm and critique. By mapping Thermomix use cases onto examples of learning with generative AI, and situating them within the ICAP and SAMR frameworks, we show how different modes of tool use can either support or undermine meaningful engagement and learning. The Thermomix metaphor underscores that the central question is not whether learners employ AI, but how such use shapes their learning processes. In doing so, we provide a conceptual lens for researchers and practitioners to critically examine - and more effectively guide - the integration of generative AI into educational practice.

Although probabilistic statements are ubiquitous, foundational disagreements persist about their understanding, as exemplified by debates between Bayesians and frequentists; moreover, it is unclear when and why acting on them actually leads to desirable outcomes. Here, we argue that every probability is the output of a prediction method, that is, it depends on both a particular way of constructing abstractions and a way of transforming them into predictions. Through this, we provide a unifying perspective on supposedly different kinds of probabilities and show that even supposedly objective ones are model-dependent. We demonstrate that when a finite calibration criterion is met, one can anticipate the distribution of utilities for a given policy and inform successful decision-making on finite sets of events. Based on the notion of prediction methods, inductive arguments, and the probability calculus, we explain the feasibility of the calibration criterion in many settings. Overall, we develop a coherent perspective on probabilities and their use, connecting key intuitions behind other interpretations along the way.

Studies of careers in science, film, music, and books report a common pattern. When a person's most successful work arrives is close to a random draw over the works they produce. How large their successes tend to be, in contrast, follows a stable, person-specific factor. We test whether this pattern holds for open-source software careers and whether it changed when generative AI coding tools arrived. From the complete public record of GitHub push events (2015-2025), we reconstruct 102.2M career works by 6.15M contributors, and for the 908k contributors whose repositories publish packages, we measure each work's impact by how many downstream packages come to depend on it. First, we find that the timing of a career's biggest hit is close to a lottery over their works, as in science and the arts, with a small, replicable lean toward early career that grows as careers get longer. Second, some coders reliably produce higher-impact work than others, but this lasting personal factor accounts for only part of why impact persists (about a fifth in our primary specification); the rest behaves like momentum, success feeding on itself for a period of time. Third, within the same contributors, this structure did not change after ChatGPT's release. The stable factor's weight grew by about as much as it grew for an earlier cohort that simply aged, and subtracting the effect of aging from the effect of generative AI puts the shift at +0.03 (95% CI [-0.22, +0.23]), indistinguishable from zero. The success pattern documented in science and the arts therefore describes open-source careers too, and it shows no detectable break across the arrival of generative AI. These results have implications for how track records on open platforms should be read and on what to expect from generative AI for the careers built on them.

Large language models (LLMs) are increasingly used for individual next-location prediction, with personal information easily added to prompts alongside mobility history. Yet the incremental predictive value of such information remains unclear. Using linked sociodemographic records and mobility traces from 5,000 Shenzhen residents, this study separates model responsiveness from predictive value. GPT-5 is the primary model, with GPT-5.5 and Claude Opus 4.6 used for replication. In 1,000 paired prediction instances, models rank 100 candidate destinations with and without age, gender, occupation and income while all other inputs are held fixed. Behavioural history raises top-1 accuracy from 5.6% to 18.5% as prior history increases from zero to six days. By contrast, sociodemographic attributes produce no detectable overall gain, although replacing correct attributes with those of another person reduces accuracy by 5.4 percentage points. Candidate construction also matters, removing distance raises accuracy by 7.7 points under proximity sampling but lowers it by 22.3 points under popularity sampling, with the reversal reproduced across all three LLMs. These findings identify behavioural history as the clearest source of incremental value and show that personal information should be evaluated under matched, explicitly specified conditions before its privacy and governance costs are justified.

On August 2, 2026, the obligations of Article 50 of the EU AI Act took effect, requiring generative AI providers to mark the content their systems produce and ensure it can be detected as AI-generated. Days later, Anthropic disclosed that every Claude model released after that date embeds a watermark based on SynthID-Text in all generated text, enabled by default with no user opt-out; Google has deployed SynthID-Text in Gemini since 2024. Users objected that the watermark degrades quality, particularly for code, that it secretly encodes identifying information, and, in mutual contradiction, that it is easily removable and inescapable; the vendor answered with assurances of unchanged quality, no identifying information, and robustness to light editing. In this work, we argue that neither the objections nor the assurances can currently be verified and that this unverifiability, rather than watermarking itself, is the substantive governance failure. We sort the contested assertions by what it would take to settle each and evaluate the open-source SynthID-Text implementation on two open-weight models, because no public tool can test the deployed systems. On prose, the measured effect of the watermark does not exceed that of changing the sampling seed. On code, the cost is three points of correctness on one model and below measurement on the other, while detection remains near chance, a limitation of detectability rather than quality. The remaining gaps trace to withheld access or missing institutions and we map each to a requirement: release of matched outputs, configuration disclosure, accredited audits, a shared evaluation protocol, and interoperable detection.