Open access

AI research for systems that carry responsibility

Articles on autonomous agents, regulated finance, digital identity, ledgers, energy markets, valuation, education and insurance. Read the latest work and follow the ideas across the journal.

How to submit

SAIJ-guex5c23zfm5Oct 2026

Multi-Agent AI as a Nested Principal-Agent Problem in Private Wealth Management

Mandate Representation and Evidence Control in Switzerland, Germany and Austria

DOI 10.5281/zenodo.23095194

arXiv 2610.02863

Walter Kurz¹, Reinhard Magg¹, Florian Kollberg², Wojtek Stricker¹, Stefan Marx², Frank Reinhardt², Velimir Dedić³

¹Swissi Institute for AI²Hochschule für Wirtschaft und Umwelt Nürtingen-Geislingen³Faculty of Information Technology and Engineering (FITI), Belgrade

A nested client–manager–AI formulation specifies separate client and manager outcomes and a constrained joint objective. Analytical constructions and simulations examine how mandate representation, clarification and evidence-bound approval affect recommendations.

Keywords:

multi-agent AI
private wealth management
nested principal-agent
mandate representation
joint objective
specification risk
suitability
premise provenance
human oversight

SAIJ-5kdnql4rsq27Jul 2026

Multi-Jurisdictional Legal Identity Assurance for Capability Gating

A Design-Science Proposal for Tiered, Reusable Identity Assurance of Natural, Juridical, and Machine Entities

DOI 10.5281/zenodo.21901241

arXiv 2610.00287

Walter Kurz

Swissi Institute for AI

Flat maximum verification charges every participant for the rarest high-risk case, and excludes those who cannot clear a bar they never needed to. This model holds the assurance state apart from the capability gate that consumes it, so identity demand follows the act and the weight of its consequences rather than mere presence.

Keywords:

identity assurance
capability gating
tiered and reusable verification
multi-jurisdictional identity
data minimisation
entity taxonomy
bitemporal reliance
design science research

SAIJ-zpd6gvtrfaunJul 2026

Credentials and Triangulated Trust Signals on a Single Accountable Identifier

A Hash-Anchored Distributed-Ledger Framework for Portable Identity across Jurisdictions

DOI 10.5281/zenodo.21901243

Walter Kurz

Swissi Institute for AI

Digital identity stays rigid while it is bound to provider accounts, mutable handles and local wallet schemes. An accountable hash-anchor tier sits above them: an inert root anchor, unlinkable profile anchors for distinct contexts, and gate-specific assurance evaluated at a point in time.

Keywords:

digital identity
verifiable credentials
accountable pseudonymity
distributed ledger
selective disclosure
identity assurance
self-sovereign identity
hash anchor

SAIJ-sh27g6sykt2kJul 2026

Identity-Staked Consensus and Collusion Resistance in Chartered Validator Sets

A Trust Model for Decentralised and Compliant Distributed Settlement Infrastructure

DOI 10.5281/zenodo.21901245

Walter Kurz

Swissi Institute for AI

Permissioned ledgers are commonly dismissed as centralised because admission is restricted. Separating permissioning from control distribution makes validator identity externally costly collateral: public legal identity, charter state, liability and audit exposure, with affiliation-aware voting caps and per-member collusion margins.

Keywords:

identity-staked consensus
permissioned ledger
proof-of-authority
validator trust model
collusion resistance
settlement infrastructure
actor assurance
threshold class coverage
ledger evidence record

SAIJ-f3jtignfyge3May 2026

Firm Valuation When AI Shapes the Business Model

A Milestone-Based Real-Options Framework for the AI Valuation Uncertainty Problem

DOI 10.5281/zenodo.21901247

arXiv 2609.24181

Walter Kurz¹, Wojtek Stricker¹, Stefan Marx², Frank Reinhardt², Florian Kollberg²

¹Swissi Institute for AI²Hochschule für Wirtschaft und Umwelt Nürtingen-Geislingen

Discounted cash flow, the IDW S 1 income approach and market multiples compress milestone probabilities, continuation options and risk shifts into opaque aggregate parameters. A milestone-gated real-options overlay decomposes that value into auditable components, with a Success Readiness Index deriving per-option probabilities from structured pairwise comparisons.

Keywords:

firm valuation
AI integration
real options
milestone-based valuation
intangible assets
AHP
multi-criteria decision analysis

SAIJ-cwo7xrcdsautMar 2026

Functional Architecture of European Electricity Trading Markets

Requirements for AI Supported Trading Systems under Regulatory Constraints

DOI 10.5281/zenodo.21901249

arXiv 2609.29108

Walter Kurz, Wojtek Stricker

Swissi Institute for AI

European electricity trading runs as a constrained multi-layer system in which legal design, exchange microstructure and network physics execute jointly across forward, day-ahead, intraday and balancing horizons. The paper specifies an AI-supported trading architecture with a permission gate on executable actions and fail-closed control logic under REMIT, MiFID II, MiFIR and EMIR.

Keywords:

EU electricity market
market coupling
NEMO topology
electricity balancing
AI trading systems
compliance-by-design

SAIJ-xz3bi3q7fwimAug 2025

Compliant AI Infrastructure for Regulated Finance

A tiered multi-agent framework with DLT audit trails for financial operations in DACH

DOI 10.5281/zenodo.21901251

arXiv 2609.27632

Walter Kurz, Reinhard Magg

Swissi Institute for AI

Regulation is treated as an orientation layer rather than a deterministic ruleset: a matrix of regulatory intent and exposure is compiled into concrete prohibitions, obligations and runtime budgets. Evidence, decisions and reason codes bind to a permissioned DAG, so a supervisor can replay how an outcome was reached and attribute failure.

Keywords:

DACH finance
regulated financial institutions
multi agent expert system
policy compiled orchestration
objective under constraints
permissioned DLT
DAG timestamping
audit trails
EU AI Act
MiFID II
DORA
GDPR
human oversight
execution gating
ESG budgets
verification and assurance

SAIJ-ddkjais6s332Aug 2025

A regulatory-compliant AI and verification system for higher education under ESG-aligned constraints

DOI 10.5281/zenodo.21901253

Walter Kurz, Michel Malara, Wojtek Stricker

Swissi Institute for AI

Two linked components for higher education: a role-specific multi-agent framework for institutional operations, and a decentralised verification layer for audit, credential authentication and tamper-evident records. GDPR, the EU AI Act, EQF, ECTS and ESG directives are encoded as structural constraints rather than checked after the fact.

Keywords:

Regulatory technology
artificial intelligence in education
multi-agent AI systems
decentralised verification
academic tokenisation
GDPR compliance
EU AI Act
digital credential infrastructure
ESG governance
UniAI
UniDVS

SAIJ-zkihhbpahsbrAug 2025

Verifiable Federated AI Infrastructure

Swiss compliant federated AI DLT network using Nash equilibrium and ESG metrics

DOI 10.5281/zenodo.21901255

Walter Kurz, Michel Malara, Velimir Dedić

Swissi Institute for AI

Centralised AI infrastructure scales, and collides with latency, auditability and energy constraints. The design separates centralised training from decentralised inference and storage across five node classes, tying a size-neutral availability floor to tiered rewards for service level, ESG performance and anti-concentration.

Keywords:

decentralised data centre
AI
Federated AI infrastructure
ESG
ESG-aware compute
Nash equilibrium
digital sovereignty
Swiss data regulation
tokenised infrastructure
verifiable AI services

SAIJ-soeptiqyucowAug 2025

Generic AI-DLT Enterprise System

Architecture and methodology for scalable domain adaptation from a unified core framework

DOI 10.5281/zenodo.21901257

Walter Kurz, Michel Malara, Velimir Dedić

Swissi Institute for AI

Compliance in AI deployments is usually applied afterwards, through prompt engineering, rather than built into the foundation. This architecture encodes regulatory, governance and ESG requirements as an objective-under-constraints problem, so every specialised agent operates within legally admissible and auditable bounds before any domain work begins.

Keywords:

Compliance-first AI
Multi-agent systems
Distributed ledger technology
Directed acyclic graph
Regulation by design
ESG integration
Domain-agnostic architecture
Deployment-agnostic architecture
Vendor-agnostic architecture
Objective-under-constraints

SAIJ-qzvrl4bwy7y2May 2025

Multi-Agent AI Architecture for Regulated Insurers

A generic AI framework under Solvency II and the AI Act in Austria and Germany

DOI 10.5281/zenodo.21901259

arXiv 2609.27636

Walter Kurz

Swissi Institute for AI

The insurer is modelled as a constrained optimisation entity under solvency, legal, ESG and operational boundaries, then decomposed into specialised agents for capital, underwriting, claims, compliance and fraud. Human-in-the-loop roles enter through tiered access control, with an orchestrator enforcing regulatory admissibility across the set.

Keywords:

Multi-Agent Systems
Enterprise AI
Insurance Firms
Solvency II
AI Act
Regulated Environments
Constrained Optimisation
Principal-Agent Theory
Nash Equilibrium
Arrow’s Risk Pooling
Austria
Germany
Institutional Design
ESG Compliance
Regulatory Architecture
Model Context Protocol (MCP)
Agent-to-Agent Protocol (A2A)
AI Governance
Algorithmic Accountability
Financial Regulation

From arXiv

Recent arXiv Articles

arXiv cs.AI

SwarmEvolve: Latent Agent Swarms for Evolutionary Discovery and Optimization

LLM-guided evolutionary search uses AI agents to improve programs for mathematical and computational problems by iteratively refining solutions and combining high-scoring methods. However, the process is often slow because the LLM generates a long textual chain of thought for each round and must repeat this...

arXiv cs.LG

User Embeddings are Superpositions of Interpretable Behavioral Modes

User embeddings are central components of modern recommendation, personalization, and behavioral modeling systems. Though these embeddings can compactly encode rich behavioral information, they are inherently dense and opaque, limiting interpretability. To address this gap, we argue that user embeddings can be...

arXiv cs.AI

Scientist World: Grow a Research World for AI-Scientists

AI-Scientists arrived, but a world for them to grow in did not. Human scientists take theirs for granted - centuries of open problems and shared ideas, waiting in the scientific literature. To AI-Scientists, that world is readable but not runnable, so we build Scientist World for them: every research paper becomes...

arXiv cs.LG

XStack: A Benchmark for ML Repo Migration

ML code should run where researchers choose, not where it was first written. Today, machine learning code mainly lives in two ecosystems, PyTorch on NVIDIA GPUs and JAX on Google TPUs. Migrating repositories between them means building and testing across frameworks and hardware. This migration is (1) a practical...

arXiv cs.AI

Same Behavior, Different Mechanism: What Finite Experiments Can Certify About Integrated Information

Interpretability, AI auditing and assessments of machine consciousness increasingly rest on mechanism-level quantities: functionals of a system's causal transition law that describe how it computes, not what it outputs. Experiments see that law only through finite, partial and noisy interventions. We ask when such...

arXiv cs.LG

Towards interpreting any residual stream direction by scaling activation inversion

To interpret a direction in a language model's residual stream, we can look for text that strongly activates it. But when such text is found by scanning a corpus, it is limited to what that corpus contains. We develop Maximally-Activating Example Metamodels (MAEMs), language models which map residual stream...

arXiv cs.AI

ReproBridge: Building Reproducibility into ML Research Artifacts

Reproducing machine learning experiments requires reconstructing procedures from information scattered across papers, released code, and supporting artifacts. For coding agents, reproduction success can therefore depend on their ability to fill in missing details, rather than on the released artifacts alone. We ask...

arXiv cs.LG

Adversarial Training with Large Step Sizes: Implicit Bias, Optimization, and Sharpness

Adversarial Training (AT) is an important method for improving model robustness. We study the dynamics of deterministic Gradient Descent (GD) in AT. Most existing works in this direction focus on the small or vanishing step sizes regime, where GD resembles a continuous-time flow. However, such step sizes can make...

arXiv cs.AI

Label-Wording Augmentation in a Typed-Decision Encoder: Gains on Unfamiliar Words, Not on Abbreviations or Codes

Typed-decision models score options that the caller supplies at request time, so their accuracy depends on label wording that the model's developer does not control. We compare two releases of a 307M-parameter multilingual cross-encoder, Ru-Jev, trained on thirteen Russian-language tasks; the later release adds an...

arXiv cs.LG

Unpacking the Compute Efficiency of Strong-to-Weak OPD against GRPO

On-policy distillation (OPD) trains a student on its own rollouts under a frozen teacher's per-token supervision. Does this dense signal reduce the training FLOPs needed to reach a target accuracy compared with reinforcement learning? We systematically compare vanilla reverse-KL OPD with Group Relative Policy...

arXiv cs.AI

Understanding AI from the Inside: The State of Mechanistic Interpretability for AI Safety

Neural networks are likely to be among the most consequential artefacts in human history. As these systems become more capable and autonomous, their potential to cause societal harm also grows, making AI safety one of the defining challenges of our time. Yet we continue to build increasingly powerful systems...

arXiv cs.LG

A Hidden Markov Preference Model for Online Prediction of Adversary Actions

A defender anticipating an attacker needs a probability distribution over the attacker's next action from information available before that action occurs. Two structures complicate this: the attacker's behavioral regime persists and can switch within an episode, and each action combines an operation with...

arXiv cs.AI

MedGraph: Maintaining Patient State with Graph Memory for Long-Term Medical Dialogue

LLM-based medical assistants that follow a patient across visits need a memory that keeps the patient's clinical state correct as it changes. We examine the errors of five memory systems on longitudinal medical dialogue and find three recurring causes: the retrieved evidence gives an outdated state, includes...

arXiv cs.LG

Data-driven discovery of intrinsic Lie point symmetries of differential equation systems

Discovering continuous symmetries directly from differential-equation data can reveal geometric structure for model reduction and scientific machine learning without symbolic determining equations or a closed-form representation of the governing system. We formulate intrinsic Lie point-symmetry discovery as an...

arXiv cs.AI

The Power of Meta Access in the Software of the Future --- Demonstrated on LLM Chat UIs and Chat AI

Chat remains the default surface for many large language model (LLM) tasks, even as agentic tools such as OpenCode excel at coding and automation. Yet the model sees the current dialogue but not the workspace around it. We explore meta access---giving the assistant a structured, permissioned view of the interface...

arXiv cs.LG

On the limitations of molecular representations

Molecular representations such as fingerprints and learned embeddings play a central role in cheminformatics and small-molecule machine learning. They serve as the primary input for a diverse array of tasks, including clustering, dataset splitting, and virtual screening, yet their evaluation is often limited to...

arXiv cs.AI

Failure Prediction and Classification in Vision-Language-Action Models Via Calibrated Latent Probing

Vision-Language-Action models (VLAs) enable general-purpose embodied intelligence, but their performance is brittle under distribution shift, where unanticipated failures can damage equipment or endanger people. Deploying VLAs in real-world systems therefore requires a runtime monitor that is timely, calibrated,...

arXiv cs.LG

Spectral and Subspace Dynamics of LoRA in Supervised Fine-Tuning and Reinforcement Learning

Low-rank adaptation (LoRA) is commonly used in both supervised fine-tuning (SFT) and reinforcement learning (RL), often under similar optimizer settings. Yet, how the different dynamics of the two stages manifest in the low-rank increment that LoRA optimizes remains unclear. Tracking the singular spectrum and...

arXiv cs.AI

Benchmarks as Software: Terminal-Bench and the Case for Continuous Benchmarks

Coding agents are among the most economically consequential deployments of LLMs to date, making reliable evaluation increasingly important. Yet even benchmarks that are carefully validated at release can become stale. As models improve, they can discover valid solutions that existing verifiers reject, while changes...

arXiv cs.LG

Conformal Prediction for Time-Dependent PDEs

Uncertainty quantification is crucial in scientific machine learning, where models inform safety-critical tasks such as flood forecasting, financial risk management, and thermal control in machines. Conformal prediction provides distribution-free coverage guarantees, but in time-dependent settings common to physics...

Recent Daily Papers

From Hugging Face

Hugging Face Daily Papers

Tools Need Not Speak in Text: Latent Context Compression for Agents

Large language model (LLM) agents typically append raw tool outputs to their context. Over multi-step interactions, long and noisy observations incur substantial context overhead and dilute the agent's attention to task-relevant information. Existing token pruning and summarization methods remain constrained by...

Hugging Face Daily Papers

Signed but Steerable: Measuring and Closing the Text-Vector Integrity Gap in Offline Retrieval-Augmented Generation

Offline retrieval-augmented generation (RAG) on edge devices is updated through signed knowledge packs that carry text and precomputed embedding vectors. A signature proves who published a pack, not that its vectors encode its text. We study this text-vector integrity gap. A verifier that recomputes vectors and...

Hugging Face Daily Papers

StepX-OPD: Rethinking Multi-Teacher On-Policy Distillation in Diffusion Models

A deployable text-to-image diffusion model must hold several capabilities at once, such as instruction following, text rendering and human preference, and must sample in a few steps. Capabilities and few-step sampling are built in separate stages: a full-step model first consolidates capabilities through...

Hugging Face Daily Papers

SwarmEvolve: Latent Agent Swarms for Evolutionary Discovery and Optimization

Hugging Face Daily Papers

OpenFrontiers: A Collection of Open Research Problems and a Benchmark for Scientific Rediscovery

Large language models (LLMs) have shown increasing promise in research-level problem solving, yet a comprehensive collection of such problems is still missing, and evaluating them on research-level problems remains challenging. We introduce OpenFrontiers, a cross-disciplinary resource that combines a structured...

Hugging Face Daily Papers

Terminal-style Mathematical Modeling Bench: The First Agentic and Verifiable Benchmark for Mathematical Modeling

Mathematical modeling turns an open-ended real-world problem into assumptions, a model, a solution, and a defended recommendation, and its answers matter only if they are both close to a credible reference and physically feasible; existing benchmarks either score solver fragments or grade reports with an LLM judge,...

Hugging Face Daily Papers

Muir: Linear Attention with Sensitivity-Allocated Gating and Matched-Filter Readout

Linear attention compresses context into a fixed-size key-value state for linear-time training and constant-memory inference. Gated delta-rule models decay this state and erase its prediction for an incoming key before writing a new value; Gated DeltaNet-2 (GDN-2) generalizes this with channel-wise decay and...

Hugging Face Daily Papers

Information Games: Strategic Crowding and Firm Repositioning in Language-Model Space

When firms respond to new market opportunities, their capacity choices alter the future discounted free cash flows available to both competitors and investors. In competitive markets, these investment decisions publicized in news articles and analyst reports, measuring the changes in capital structure, product mix,...

Hugging Face Daily Papers

Reducing Representational Interference in Retrieval Embeddings with Trainable Prefix Tokens

Embedding models compress variable-length text into fixed-size vector representations, but this compression can entangle semantic content with residual token-level information, such as artifacts arising from attention sink behavior. We propose a simple method that prepends trainable [DUMP] tokens to the input...

Hugging Face Daily Papers

Hard Sample-Conditioned Decorrelated Attention for Robust Face Recognition

Multi-head attention in Vision Transformers (ViTs) allows independent attention patterns across heads, but without explicit regularization, heads may learn redundant patterns rather than complementary roles. Correlation-based penalties encourage head disentanglement, yet apply uniform regularization across all...

Hugging Face Daily Papers

On the Opportunities and Challenges of Prefill-to-Autoregressive Cross-Model Transfer

Prefill-to-Autoregressive Cross-model Transfer (PACT), a form of latent communication between language models, combines a larger model's prompt representations with a smaller model's efficient generation, offering a promising way to balance capability and inference cost. However, existing systems provide limited...

Hugging Face Daily Papers

Towards Real-Time Full-Waveform LiDAR Transformers via Intensity-Guided Token Reduction and Physics-Aware Augmentation

LiDAR is a critical sensor for autonomous driving and robotics, yet it remains vulnerable to adverse conditions such as fog, rain, transparent objects, and low-reflectance surfaces. Full-waveform LiDAR (FWL) addresses these limitations by capturing the complete return waveform as a histogram, preserving rich...

Hugging Face Daily Papers

SynthFaces: Balanced Large Scale Human Dataset

The pursuit of fair and equitable AI in human-centric tasks is critically hindered by the inherent biases and privacy risks of real-world datasets, which offer imbalanced representations of identity and often contain sensitive personal information. Consequently, state-of-the-art generative models trained on this...

Hugging Face Daily Papers

Decoupled Predictive Decision Making: A Shared Conditional Predictor for Agents

Agent decisions depend on empirical relationships that deployment observations reveal and task descriptions alone may leave unresolved. We introduce Decoupled Predictive Decision Making (DPDM), an architecture that formulates heterogeneous agent decisions as conditional prediction with a shared pretrained learner....

Hugging Face Daily Papers

SAE Interprets MoE: Sparse AutoEncoders Reveal That MoE Experts Exhibit Human-Interpretable Fine-Grained Specialization

Mixture-of-Experts (MoE) models route each token to a few of many experts, yet what an individual expert is responsible for remains unsettled. Existing analyses count how often an expert is selected on domain-labeled corpora, an instrument that fixes the taxonomy in advance, stays coarse, and records selection...

Hugging Face Daily Papers

HeLM: Hierarchical Explicit Language Memory for Non-Markovian Manipulation

Recent advances in Vision-Language-Action (VLA) models have enabled fine manipulation capabilities, yet performing tasks that require working memory remains a key challenge. Most existing VLA architectures operate under a Markovian assumption, mapping only the current observation to an action, which leads to...

Hugging Face Daily Papers

TailSteer: Trajectory-Tail Steering for Drift-Resistant Multi-turn Image Editing

Strong single-turn editing performance does not guarantee reliable multi-turn interaction: as edited outputs are repeatedly reused, small unintended changes can accumulate into severe visual drift. This long-term drift is often mitigated through training on multi-turn data or cross-turn rollouts, potentially...

Hugging Face Daily Papers

From Autonomous Agents to Cognitive Symbionts: Toward AI-Augmented Human Intelligence

For a decade, progress in artificial intelligence has been measured by how much a machine can do on its own. Large foundation models now write code and run multistep tasks, and personal agents act on a person's behalf. Yet people still forget what they once knew, miss what they are not looking at, fail to find the...

Hugging Face Daily Papers

EvoMemState: Evolving Memory State for Personal Agents

Long-term memory enables large language model (LLM) agents to provide continuous personalized assistance by retaining user information across interactions. Existing systems support this process through fact extraction, organization, updating, and retrieval, but largely maintain memories as independent facts. This...

Hugging Face Daily Papers

Decode Before You Hear: Speculative Listening for Chunk-Interleaved Full-Duplex Speech Language Models

Chunk-interleaved full-duplex speech language models listen and speak in fixed time chunks, and each chunk must be processed before the next one arrives. Their workload is asymmetric across chunks: listen chunks leave compute idle while awaiting audio, and speak and tool-call chunks require serial generation, which...