Research Papers

This paper introduces a performative scenario optimization framework for decision-dependent chance-constrained problems. Unlike classical stochastic optimization, we account for the feedback loop where decisions actively shape the underlying data-generating process. We define performative solutions as self-consistent equilibria and establish their existence using Kakutani's fixed-point theorem. To ensure computational tractability without requiring an explicit model of the environment, we propose a model-free, scenario-based approximation that alternates between data generation and optimization. Under mild regularity conditions, we prove that a stochastic fixed-point iteration, equipped with a logarithmic sample size schedule, converges almost surely to the unique performative solution. The effectiveness of the proposed framework is demonstrated through an emerging AI safety application: deploying performative guardrails against Large Language Model (LLM) jailbreaks. Numerical results confirm the co-evolution and convergence of the guardrail classifier and the induced adversarial prompt distribution to a stable equilibrium.

Medical device regulators in the United States(FDA), China (NMPA), and Europe (EU MDR) all use the language of risk, but classify devices through structurally different mechanisms. Whether these apparently shared concepts carry transferable classificatory signal across jurisdictions remains unclear. We test this by reframing explainable AI as an empirical probe of cross-jurisdictional regulatory overlap. Using 141,942 device records, we derive seven EU MDR risk factors, including implantability, invasiveness, and duration of use, and evaluate their contribution across a three-by-three transfer matrix. Under a symmetric extraction pipeline designed to remove jurisdiction-specific advantages, factor contribution is negligible in all jurisdictions, indicating that clean cross-jurisdictional signal is at most marginal. Under jurisdiction specific pipelines, a modest gain appears only in the EU MDR-to-NMPA direction, but sensitivity analyses show that this effect is weak, context-dependent, and partly confounded by extraction and representation choices. Reverse direction probes show strong asymmetry: FDA-derived factors do not transfer meaningfully in any direction, and NMPA-derived factors do not carry signal back to EU MDR. Zero-shot transfer further fails on EU MDR Class I, consistent with a mismatch between residual and positional class definitions. Overall, cross-jurisdictional transfer is sparse, asymmetric, and weak. Shared regulatory vocabulary does not, under this operationalisation, translate into strong portable classification logic. The findings challenge a common assumption in cross-jurisdictional regulatory AI and show how explainable AI can be used to measure, rather than assume, regulatory overlap.

We propose a stylized model of human-AI collaboration that isolates a mechanism we call the novelty bottleneck: the fraction of a task requiring human judgment creates an irreducible serial component analogous to Amdahl's Law in parallel computing. The model assumes that tasks decompose into atomic decisions, a fraction $ν$ of which are "novel" (not covered by the agent's prior), and that specification, verification, and error correction each scale with task size. From these assumptions, we derive several non-obvious consequences: (1) there is no smooth sublinear regime for human effort it transitions sharply from $O(E)$ to $O(1)$ with no intermediate scaling class; (2) better agents improve the coefficient on human effort but not the exponent; (3) for organizations of n humans with AI agents, optimal team size decreases with agent capability; (4) wall-clock time achieves $O(\sqrt{E})$ through team parallelism but total human effort remains $O(E)$; and (5) the resulting AI safety profile is asymmetric -- AI is bottlenecked on frontier research but unbottlenecked on exploiting existing knowledge. We show these predictions are consistent with empirical observations from AI coding benchmarks, scientific productivity data, and practitioner reports. Our contribution is not a proof that human effort must scale linearly, but a framework that identifies the novelty fraction as the key parameter governing AI-assisted productivity, and derives consequences that clarify -- rather than refute -- prevalent narratives about intelligence explosions and the "country of geniuses in a data center."

National and international policy efforts increasingly promote AI literacy in K--12 education, yet access to AI tools alone is insufficient to ensure equitable and responsible participation, particularly in under-resourced contexts. Without designs that integrate ethical reasoning, human support, and opportunities for judgment, AI initiatives risk reinforcing inequities and preparing students to use powerful technologies without critically evaluating their societal consequences. To address this gap, we developed a human-centered, college credit--bearing course on Responsible and Ethical AI for students attending Title I and Title I--eligible high schools, implemented in partnership with the National Education Opportunity Network (NEON). The bichronous course integrates foundational AI concepts with ethical and socio-technical reasoning through asynchronous instruction, near-peer mentorship, and synchronous, discussion-based instruction. In its inaugural year, nearly 180 students from 12 U.S. schools completed the course, with a 97.8% completion rate. Using end-of-course survey data from students, co-teachers, and teaching fellows, this study examines academic agency, confidence with college-level expectations, critical engagement with responsible AI, and perceived post-secondary trajectory expansion. Results indicate that students were challenged to apply learning and reason about ethical tradeoffs, while educators reported high engagement, rigor, and meaningfulness relative to typical high school coursework. Overall, findings highlight the importance of human connections in advancing equitable Artificial Intelligence in Education and support ethical judgment as a core learning outcome alongside AI literacy.

We prove that platform-deterministic inference is necessary and sufficient for trustworthy AI. We formalize this as the Determinism Thesis and introduce trust entropy to quantify the cost of non-determinism, proving that verification failure probability equals 1 - 2^{-H_T} exactly. We prove a Determinism-Verification Collapse: verification under determinism requires O(1) hash comparison; without it, the verifier faces an intractable membership problem. IEEE 754 floating-point arithmetic fundamentally violates the determinism requirement. We resolve this by constructing a pure integer inference engine that achieves bitwise identical output across ARM and x86. In 82 cross-architecture tests on models up to 6.7B parameters, we observe zero hash mismatches. Four geographically distributed nodes produce identical outputs, verified by 356 on-chain attestation transactions. Every major trust property of AI systems (fairness, robustness, privacy, safety, alignment) presupposes platform determinism. Our system, 99,000 lines of Rust deployed across three continents, establishes that AI trust is a question of arithmetic.

We present the Consilium Protocol, a Byzantine Fault Tolerance-derived architecture for structured multi-model AI deliberation that treats inter-model disagreement as epistemic signal rather than error. The protocol assigns engineered cognitive personas to language models -- separating what a model is from how it reasons -- and introduces an In-Sample/Out-of-Sample validation framework adapted from quantitative finance to distinguish training-data consensus from empirically grounded conclusions. Across 1,478 deliberation sessions spanning 32 topics in 10 domain categories, we demonstrate that (1) the cognitive persona, not the underlying model, determines epistemic behavior: free edge-inference models costing 0.0002 USD per batch produced comparable analytical output to frontier models costing 10.69 USD; (2) RLHF alignment training creates measurable, domain-specific epistemic blind spots -- contested policy topics exhibit 12.3 percentage points less adversarial challenge than settled science topics, and AI safety topics show asymmetric bias ($Δ$=11.6%) where models challenge claims that AI is dangerous far more vigorously than claims that AI risk is overstated; (3) the protocol exhibits no directional bias of its own (immigration $Δ$=2.3%, renewables $Δ$=1.2%); and (4) out-of-sample evidence retrieval validated 239 claims with 100% evidence retrieval and surfaced 167 blind-spot discoveries invisible to training-data deliberation. Run-to-run reproducibility across randomized model$\times$persona assignments averages $\pm$2.2% standard deviation. Total cost for the complete battery including all overhead: 217 USD. We release the protocol specification under MIT license to enable independent verification.

The first generation of students is learning to program alongside GenAI (Generative Artificial Intelligence) tools, raising questions about how young learners critically engage with them and perceive ethical responsibilities. While prior research has focused on university students or developers, little is known about secondary school novices, who represent the next cohort of software engineers. To address this gap, we conducted an exploratory study with 84 German secondary school students aged 16-19 attending software development workshops. We examined their critical thinking practices in AI-assisted programming, perceptions of AI ethics and responsibility, and gender-related differences in their views. Our results reveal an AI paradox: students demonstrate strong ethical reasoning and awareness about AI, yet many report integrating AI-generated code without a thorough understanding of it. The majority of our cohort attributed significant responsibility for AI practices to politics and corporations, potentially reflecting Germany's cultural context, with its strict regulations and data privacy discourse. Boys reported more frequent and experimental use of AI-assisted programming, whereas girls expressed greater scepticism and emphasised peer collaboration over GenAI assistance. Our findings highlight the importance of culturally responsive software engineering education that strengthens critical AI literacy in AI-assisted programming by linking ethics to concrete code artefacts and preparing young learners for this AI-driven software landscape.

The safety of autonomous AI agents is increasingly recognized as a critical open problem. As agents transition from passive text generators to active actors capable of executing shell commands, modifying files, calling APIs, and browsing the web, the consequences of unsafe or adversarially manipulated behavior become immediate and tangible. Existing AI safety work has focused primarily on model alignment and input filtering, but these approaches do not address what happens at the moment an agent's intent becomes a real action on a real machine. This gap is especially acute in local environments, where developers run agents against their own filesystems, credentials, and infrastructure with little runtime control. This paper introduces AgentWall, a runtime safety and observability layer for local AI agents. AgentWall intercepts every proposed agent action before it reaches the host environment, evaluates it against an explicit declarative policy, requires human approval for sensitive operations, and records a complete execution trail for audit and replay. It is implemented as a policy-enforcing MCP proxy and native OpenClaw plugin, working across Claude Desktop, Cursor, Windsurf, Claude Code, and OpenClaw with a single install command. We present the design, architecture, threat model, and policy model of AgentWall, and demonstrate 92.9% policy enforcement accuracy with sub-millisecond overhead across 14 benchmark tests. AgentWall is open-source at https://github.com/agentwall/Agentwall.

Shaping Scientific Explanations to Expert Perspectives with Persona-Conditioned Reinforcement Learning

Susana Nunes, Tiago Guerreiro, Catia Pesquita

Explainable AI is increasingly important to scientific discovery. However, existing methods largely ignore that explanation quality is not universal: experts differ in how they assess evidence, prioritize mechanisms, and construct explanatory narratives. We introduce perspective-conditioned explanations, a framework for adapting explanation generation to epistemic variation in expert judgment. Using knowledge graph reasoning paths in drug discovery, we show that preferences organize into coherent epistemic perspectives that can be captured by agentic personas, representations of how experts evaluate explanations. Persona-aligned rewards then guide reinforcement learning-based explanation generation without large-scale expert supervision. Expert user studies show that perspective-conditioned explanations are preferred over general-purpose explanations and improve perceived relevance and validity. Moreover, they match or exceed state-of-the-art predictive performance and reduce expert feedback time by two orders of magnitude. Together, these findings demonstrate that explanation quality is perspective-dependent and that modeling this variation enables scalable and human-aligned explanation generation for scientific discovery.

Teachers face growing pressure to integrate AI tools into their classrooms, yet are rarely positioned as agentic decision-makers in this process. Understanding the criteria teachers use to evaluate AI tools, and the conditions that support such reasoning, is essential for responsible AI integration. We address this gap through a two-day national summit in which 61 U.S. K-12 mathematics educators developed personal rubrics for evaluating AI classroom tools. The summit was designed to support deliberative sensemaking, a process we conceptualize by integrating Technological Pedagogical Content Knowledge (TPACK) with deliberative agency. Teachers generated over 200 criteria - initial articulations spanning four higher-order themes (Practical, Equitable, Flexible, and Rigorous) - that addressed both AI outputs and the process of using AI. Criteria contained productive tensions (e.g., personalization versus fairness, adaptability versus efficiency), and the vast majority framed AI as an assistant rather than a coaching tool for professional learning. Analysis of surveys, interviews, and summit discussions revealed five mechanisms supporting deliberative sensemaking: time and space for deliberation, artifact-centered sensemaking, collaborative reflection through diverse viewpoints, knowledge-building, and psychological safety. Across these mechanisms, TPACK and agency operated in a mutually reinforcing cycle - knowledge-building enabled more grounded evaluative judgment, while the act of constructing criteria deepened teachers' understanding of tools. We discuss implications for edtech developers seeking practitioner input, school leaders making adoption decisions, educators and professional learning designers, and researchers working to elicit teachers' evaluative reasoning about rapidly evolving technologies.

This paper addresses the critical challenge of mesa-optimization in AI safety by providing a formal definition of agency and a framework for its analysis. Agency is conceptualized as a Continuous Representation of accumulated experience that achieves autopoiesis through a dynamic balance between curiosity (minimizing prediction error to ensure non-computability and novelty) and empowerment (maximizing the control channel's information capacity to ensure subjectivity and goal-directedness). Empirical evidence suggests that this active inference-based model successfully accounts for classical instrumental goals, such as self-preservation and resource acquisition. The analysis demonstrates that the proposed agency function is smooth and convex, possessing favorable properties for optimization. While agentic functions occupy a vanishingly small fraction of the total abstract function space, they exhibit logarithmic convergence in sparse environments. This suggests a high probability for the spontaneous emergence of agency during the training of modern, large-scale models. To quantify the degree of agency, the paper introduces a metric based on the distance between the behavioral equivalents of a given system and an "ideal" agentic function within the space of canonicalized rewards (STARC). This formalization provides a concrete apparatus for classifying and detecting mesa-optimizers by measuring their proximity to an ideal agentic objective, offering a robust tool for analyzing and identifying undesirable inner optimization in complex AI systems.

This paper presents ARYA, a composable, physics-constrained, deterministic world model architecture built on five foundational principles: nano models, composability, causal reasoning, determinism, and architectural AI safety. We demonstrate that ARYA satisfies all canonical world model requirements, including state representation, dynamic prediction, causal and physical awareness, temporal consistency, generalization, learnability, and planning and control. Unlike monolithic foundation models, the ARYA foundation model implements these capabilities through a hierarchical system-of-system-of-systems of specialized nano models, orchestrated by AARA (ARYA Autonomous Research Agent), an always-on cognitive daemon that executes a continuous sense-decide-act-learn loop. The nano model architecture provides linear scaling, sparse activation, selective untraining, and sub-20-second training cycles, resolving the traditional tension between capability and computational efficiency. A central contribution is the Unfireable Safety Kernel: an architecturally immutable safety boundary that cannot be disabled or circumvented by any system component, including its own self-improvement engine. This is not a social or ethical alignment statement; it is a technical framework ensuring human control persists as autonomy increases. Safety is an architectural constraint governing every operation, not a policy layer applied after the fact. We present formal alignment between ARYA's architecture and canonical world model requirements, and report summarizing its state-of-the-art performance across 6 of 9 competitive benchmarks head-to-head with GPT-5.2, Opus 4.6, and V-JEPA-2. All with zero neural network parameters, across seven active industry domain nodes spanning aerospace, pharma manufacturing, oil and gas, smart cities, biotech, defense, and medical devices.