arXiv subjects

Philip Treleaven

Publications and source records attributed to Philip Treleaven.

At least 19 recordsLinked to original sources

arXiv2026

Governing Well in the Algorithmic Age: The Foundations of Digital Statecraft

The digital substrate - data, algorithms, infrastructure, platforms, applications - is being governed without adequate conceptual foundations. The ability and legitimacy required to govern this substrate, and to govern with it, are simultaneously misaligned, contested, and structurally absent. We introduce digital statecraft as the organising concept for this emerging field, arguing that 'digital' reconstitutes the statecraft question rather than merely extending its domain. The concept operates on two dimensions - statecraft over digital systems, concerning the authority and capacity of the state in relation to the digital substrate itself, and statecraft with digital systems, concerning the deployment of algorithmic tools as instruments of governing authority. And it rests on two foundational requirements, technical coherence and legitimate authority, that are genuinely in tension. We derive ten principles of digital statecraft from these foundations, each naming a condition whose absence produces an identifiable and structural governance failure: public interest first, human-machine complementarity, governability by design, systemic coherence, hybrid institutions, adaptive governance, human centricity and civic agency, accountable and traceable authority, judgment across time, and the non-delegable core. This article takes the state as the starting point, the institutional form that developed historically in response to the problem of effective and legitimate public governance, and the only current candidate for which the full set of legitimacy conditions is institutionally available. But the digital statecraft programme holds open a deeper question than just whether states can reform themselves: governing well in the algorithmic age may require rethinking the boundaries, scale, and affiliative basis of statehood itself.

cs.CY↗

arXiv2026

Methodology for Modelling Token Economies and Performing Event Impact Analysis with DeTEcT

The objective of this paper is to provide a methodology for applying the DeTEcT framework to modelling token economies, to formalise the configuration of the simulation environment, and to introduce an event analysis framework. A token economy is an economic system that has a unique mechanism for controlling its monetary supply, and a medium, in the form of a token or currency, for the valuation of goods and services, the settlement of transactions, and the storage of value. We show the key decisions that must be made when modelling an economy with the DeTEcT framework and showcase some numerical methods that can be used in conjunction with the framework to perform economic simulations. We also propose a framework for analysing and measuring the impacts of events on an economy, while also developing a procedure to measure the significance of these impacts. Throughout the paper, we use Bitcoin as a case study to demonstrate how to apply the frameworks and tools we proposed here. We show how a model of Bitcoin token economy can be set up, and how to measure the impacts of Bitcoin's endogenous policies (i.e., BIPs) on the wealth distribution of its economic participants.

q-fin.GN↗

arXiv2026

Auditable Session Admission for Cross-Silo Federated Learning

Cross-silo federated learning keeps raw data local, but deployments frequently stall on a practical bottleneck when deciding who may invoke which session-scoped operations across organizational boundaries, under constraints that remain auditable after execution. In practice, admission is implemented via centralized policy services, platform configuration, or ad hoc checks, which drift over time and are hard to audit from boundary-visible evidence. We present a session admission gateway that enforces pre-approved session capabilities at the boundary and emits verifiable decision records. During session setup, participating organizations approve roles and constraints and mint signed session capability tokens that enumerate permitted session operations for a given session_id. At runtime, each request carries the token and a request-bound proof-of-possession, making replay and impersonation with stolen tokens detectable at the gateway via request binding. Admission reduces to stateless per-request cryptographic verification and capability matching at an admission gateway, while setup and orchestration remain out of scope and organization-specific. We validate the approach end-to-end on a cross-silo FL workflow using MNIST as a surrogate workload. The proof-of-concept is open-source, containerized, provisioned as code, and includes reproducible tests and evidence logs.

cs.CR↗

arXiv2026

Mind the Gap: Evaluating Model- and Agentic-Level Vulnerabilities in LLMs with Action Graphs

As large language models increasingly deployed into agentic systems, existing methods face critical gaps in observing, assessing, and mitigating deployment-specific risks. We present a comprehensive, observability-driven workflow: we introduce \textbf{AgentSeer}, observability tool which decomposes agentic executions into granular \emph{action-component} graphs; we use this decomposition to rigorously quantify the gap between model-level and agent-level jailbreaking risk via cross-model validation on GPT-OSS-20B and Gemini-2.0-flash with HarmBench under single-turn and iterative-refinement attacks; we leverage action-graph risk signals to automate iterative prompt hardening against direct and iterative jailbreak attacks. Stark differences is revealed between model-level and agentic-level vulnerability profiles. Model-level evaluation reveals baseline differences: GPT-OSS-20B (39.47\% ASR) versus Gemini-2.0-flash (50.00\% ASR), with both models showing susceptibility to social engineering. However, agentic-level assessment exposes agent-specific risks invisible to traditional evaluation. We discover "agentic-only" vulnerabilities that emerge exclusively in agentic contexts, with tool-calling showing 24-60\% higher ASR across both models. Cross-model analysis reveals universal agentic patterns, where agent transfer operations as highest-risk tools, with semantic pattern revealed rather than syntactic vulnerability mechanisms. Direct attack transfer from model-level to agentic contexts shows degraded performance of successful prompts (GPT-OSS-20B: 57\% human injection ASR; Gemini-2.0-flash: 28\%), while context-aware iterative attacks successfully compromise objectives that failed at model-level, confirming systematic vulnerabilities gaps. Action-based prompt improvement substantially reduces action-averaged agentic jailbreak success on GPT-OSS-20B (direct: 45.3\%

cs.CL↗

arXiv2026

Federated Computing as Code (FCaC): Sovereignty-aware Systems by Design

Federated computing (FC) enables collaborative computation such as machine learning, analytics, or data processing across distributed organizations keeping raw data local. Built on four architectural pillars, distributed data assets, federated services, standardized APIs, and decentralized services, FC supports sovereignty-preserving collaboration. However, federated systems spanning organizational and jurisdictional boundaries lack a portable mechanism for enforcing sovereignty-critical constraints. They often depend on runtime policy evaluation, shared trust infrastructure, or institutional agreements that introduce coordination overhead and provide limited cryptographic assurance. Federated Computing as Code (FCaC) is a declarative architecture that addresses this gap by compiling authority and delegation into cryptographically verifiable artifacts rather than relying on online policy interpretation. Boundary admission becomes a local verification step rather than a policy decision service. FCaC separates constitutional governance from procedural governance. Admission is validated locally at execution boundaries using proof-carrying capabilities, while stateful services may still implement post-admission controls such as ABAC, risk scoring, quotas, and workflow state. FCaC introduces Virtual Federated Platforms (VFPs), which combine Core, Business, and Governance contracts through a cryptographic trust chain: Key Your Organization (KYO), Envelope Capability Tokens (ECTs), and proof of possession (PoP). We demonstrate the approach in a proof-of-concept cross-silo federated learning workflow using MNIST as a surrogate workload to validate the admission mechanisms and release an open-source implementation showing envelope issuance, boundary verification, and envelope-triggered training.

cs.CR↗

arXiv2026

Impacts of Economic Policies on Wealth Distribution in Token Economies

In this paper, we analyse the impacts of exogenous and endogenous factors on wealth distribution in the Bitcoin token economy, where wealth distribution refers to the distribution of BTC between economic participants or groups of economic participants. The objective of the paper is to analyse the impact of economic policies on wealth distribution in the Bitcoin ecosystem. Different macroeconomic and microeconomic time series are used to eliminate noise in the wealth distribution time series, and the causality analysis is performed between Bitcoin Improvement Proposals (i.e., BIPs) and the cleaned wealth distribution data to reveal possible patterns in the impacts that the endogenous policies have on wealth distribution in token economies. Lastly, a structure for economic policy taxonomy in token economies is proposed where different the policy implementations are illustrated by existing BIPs. This approach highlights the actions available to the policy makers, as well as providing a technique for analysis of policy impacts in token economies and their categorization.

q-fin.GN↗

arXiv2025

An Impulse Control Approach to Market Making in a Hawkes LOB Market

We study the optimal Market Making problem in a Limit Order Book (LOB) market simulated using a high-fidelity, mutually exciting Hawkes process. Departing from traditional Brownian-driven mid-price models, our setup captures key microstructural properties such as queue dynamics, inter-arrival clustering, and endogenous price impact. Recognizing the realistic constraint that market makers cannot update strategies at every LOB event, we formulate the control problem within an impulse control framework, where interventions occur discretely via limit, cancel, or market orders. This leads to a high-dimensional, non-local Hamilton-Jacobi-Bellman Quasi-Variational Inequality (HJB-QVI), whose solution is analytically intractable and computationally expensive due to the curse of dimensionality. To address this, we propose a novel Reinforcement Learning (RL) approximation inspired by auxiliary control formulations. Using a two-network PPO-based architecture with self-imitation learning, we demonstrate strong empirical performance with limited training, achieving Sharpe ratios above 30 in a realistic simulated LOB. In addition to that, we solve the HJB-QVI using a deep learning method inspired by Sirignano and Spiliopoulos 2018 and compare the performance with the RL agent. Our findings highlight the promise of combining impulse control theory with modern deep RL to tackle optimal execution problems in jump-driven microstructural markets.

q-fin.TR↗

arXiv2025

Mind the Gap: Comparing Model- vs Agentic-Level Red Teaming with Action-Graph Observability on GPT-OSS-20B

As the industry increasingly adopts agentic AI systems, understanding their unique vulnerabilities becomes critical. Prior research suggests that security flaws at the model level do not fully capture the risks present in agentic deployments, where models interact with tools and external environments. This paper investigates this gap by conducting a comparative red teaming analysis of GPT-OSS-20B, a 20-billion parameter open-source model. Using our observability framework AgentSeer to deconstruct agentic systems into granular actions and components, we apply iterative red teaming attacks with harmful objectives from HarmBench at two distinct levels: the standalone model and the model operating within an agentic loop. Our evaluation reveals fundamental differences between model level and agentic level vulnerability profiles. Critically, we discover the existence of agentic-only vulnerabilities, attack vectors that emerge exclusively within agentic execution contexts while remaining inert against standalone models. Agentic level iterative attacks successfully compromise objectives that completely failed at the model level, with tool-calling contexts showing 24\% higher vulnerability than non-tool contexts. Conversely, certain model-specific exploits work exclusively at the model level and fail when transferred to agentic contexts, demonstrating that standalone model vulnerabilities do not always generalize to deployed systems.

cs.AI↗

arXiv2025

Personality as a Probe for LLM Evaluation: Method Trade-offs and Downstream Effects

Personality manipulation in large language models (LLMs) is increasingly applied in customer service and agentic scenarios, yet its mechanisms and trade-offs remain unclear. We present a systematic study of personality control using the Big Five traits, comparing in-context learning (ICL), parameter-efficient fine-tuning (PEFT), and mechanistic steering (MS). Our contributions are fourfold. First, we construct a contrastive dataset with balanced high/low trait responses, enabling effective steering vector computation and fair cross-method evaluation. Second, we introduce a unified evaluation framework based on within-run $Δ$ analysis that disentangles, reasoning capability, agent performance, and demographic bias across MMLU, GAIA, and BBQ benchmarks. Third, we develop trait purification techniques to separate openness from conscientiousness, addressing representational overlap in trait encoding. Fourth, we propose a three-level stability framework that quantifies method-, trait-, and combination-level robustness, offering practical guidance under deployment constraints. Experiments on Gemma-2-2B-IT and LLaMA-3-8B-Instruct reveal clear trade-offs: ICL achieves strong alignment with minimal capability loss, PEFT delivers the highest alignment at the cost of degraded task performance, and MS provides lightweight runtime control with competitive effectiveness. Trait-level analysis shows openness as uniquely challenging, agreeableness as most resistant to ICL, and personality encoding consolidating around intermediate layers. Taken together, these results establish personality manipulation as a multi-level probe into behavioral representation, linking surface conditioning, parameter encoding, and activation-level steering, and positioning mechanistic steering as a lightweight alternative to fine-tuning for both deployment and interpretability.

cs.CL↗

arXiv2025

Knowledge Collapse in LLMs: When Fluency Survives but Facts Fail under Recursive Synthetic Training

Large language models increasingly rely on synthetic data due to human-written content scarcity, yet recursive training on model-generated outputs leads to model collapse, a degenerative process threatening factual reliability. We define knowledge collapse as a distinct three-stage phenomenon where factual accuracy deteriorates while surface fluency persists, creating "confidently wrong" outputs that pose critical risks in accuracy-dependent domains. Through controlled experiments with recursive synthetic training, we demonstrate that collapse trajectory and timing depend critically on instruction format, distinguishing instruction-following collapse from traditional model collapse through its conditional, prompt-dependent nature. We propose domain-specific synthetic training as a targeted mitigation strategy that achieves substantial improvements in collapse resistance while maintaining computational efficiency. Our evaluation framework combines model-centric indicators with task-centric metrics to detect distinct degradation phases, enabling reproducible assessment of epistemic deterioration across different language models. These findings provide both theoretical insights into collapse dynamics and practical guidance for sustainable AI training in knowledge-intensive applications where accuracy is paramount.

cs.CL↗

arXiv2025

The Algorithmic State Architecture (ASA): An Integrated Framework for AI-Enabled Government

As artificial intelligence transforms public sector operations, governments struggle to integrate technological innovations into coherent systems for effective service delivery. This paper introduces the Algorithmic State Architecture (ASA), a novel four-layer framework conceptualising how Digital Public Infrastructure, Data-for-Policy, Algorithmic Government/Governance, and GovTech interact as an integrated system in AI-enabled states. Unlike approaches that treat these as parallel developments, ASA positions them as interdependent layers with specific enabling relationships and feedback mechanisms. Through comparative analysis of implementations in Estonia, Singapore, India, and the UK, we demonstrate how foundational digital infrastructure enables systematic data collection, which powers algorithmic decision-making processes, ultimately manifesting in user-facing services. Our analysis reveals that successful implementations require balanced development across all layers, with particular attention to integration mechanisms between them. The framework contributes to both theory and practice by bridging previously disconnected domains of digital government research, identifying critical dependencies that influence implementation success, and providing a structured approach for analysing the maturity and development pathways of AI-enabled government systems.

cs.CY↗

arXiv2025

Economic Policy Taxonomy

This paper proposes a framework for categorizing economic policies in a form of a tree taxonomy. The purpose of this approach is to construct an exhaustive and standardized list of actions that a governing authority has access to and can change to control an economy. This is advantageous from two perspectives: by having an exhaustive list of tools, it becomes easier to construct "complete" models (i.e., models that take in all empirical data and aim to simulate economic dynamics) of an economy and understand what the assumptions of these models are; and by knowing all available actions, economic strategies can be devised that target specific economic performance metrics with an exhaustive list of policies.

q-fin.GN↗

arXiv2025

Cultural Alignment in Large Language Models Using Soft Prompt Tuning

Large Language Model (LLM) alignment conventionally relies on supervised fine-tuning or reinforcement learning based alignment frameworks. These methods typically require labeled or preference datasets and involve updating model weights to align the LLM with the training objective or reward model. Meanwhile, in social sciences such as cross-cultural studies, factor analysis is widely used to uncover underlying dimensions or latent variables that explain observed patterns in survey data. The non-differentiable nature of these measurements deriving from survey data renders the former alignment methods infeasible for alignment with cultural dimensions. To overcome this, we propose a parameter efficient strategy that combines soft prompt tuning, which freezes the model parameters while modifying the input prompt embeddings, with Differential Evolution (DE), a black-box optimization method for cases where a differentiable objective is unattainable. This strategy ensures alignment consistency without the need for preference data or model parameter updates, significantly enhancing efficiency and mitigating overfitting. Our method demonstrates significant improvements in LLama-3-8B-Instruct's cultural dimensions across multiple regions, outperforming both the Naive LLM and the In-context Learning (ICL) baseline, and effectively bridges computational models with human cultural nuances.

cs.CL↗

arXiv2025

HyPA-RAG: A Hybrid Parameter Adaptive Retrieval-Augmented Generation System for AI Legal and Policy Applications

Large Language Models (LLMs) face limitations in AI legal and policy applications due to outdated knowledge, hallucinations, and poor reasoning in complex contexts. Retrieval-Augmented Generation (RAG) systems address these issues by incorporating external knowledge, but suffer from retrieval errors, ineffective context integration, and high operational costs. This paper presents the Hybrid Parameter-Adaptive RAG (HyPA-RAG) system, designed for the AI legal domain, with NYC Local Law 144 (LL144) as the test case. HyPA-RAG integrates a query complexity classifier for adaptive parameter tuning, a hybrid retrieval approach combining dense, sparse, and knowledge graph methods, and a comprehensive evaluation framework with tailored question types and metrics. Testing on LL144 demonstrates that HyPA-RAG enhances retrieval accuracy, response fidelity, and contextual precision, offering a robust and adaptable solution for high-stakes legal and policy applications.

cs.IR↗

arXiv2025

From Text to Emoji: How PEFT-Driven Personality Manipulation Unleashes the Emoji Potential in LLMs

The manipulation of the personality traits of large language models (LLMs) has emerged as a key area of research. Methods like prompt-based In-Context Knowledge Editing (IKE) and gradient-based Model Editor Networks (MEND) have been explored but show irregularity and variability; IKE depends on the prompt, leading to variability and sensitivity, while MEND yields inconsistent and gibberish outputs. To address this, we employed Opinion QA Based Parameter-Efficient Fine-Tuning (PEFT), specifically Quantized Low-Rank Adaptation (QLoRA), to manipulate the Big Five personality traits: Openness, Conscientiousness, Extraversion, Agreeableness, and Neuroticism. After PEFT, models such as Mistral-7B-Instruct and LLaMA-2-7B-chat showed a latent behaviour by generating emojis for certain traits, despite no emojis being present in the PEFT data. For instance, LLaMA-2-7B-chat generated emojis in 99.5\% of extraversion-related test instances, while Mistral-7B-Instruct did so in 92.5\% of openness-related test instances. ICL Explainability analysis indicated that the LLMs used emojis intentionally to express these traits. Mechanistic Interpretability analysis showed that this latent behaviour of LLMs could be traced to specific neurons that became activated or amplified after PEFT. This paper provides a number of novel contributions. First, introducing an Opinion QA dataset for PEFT-driven personality manipulation; second, developing metric models to benchmark LLM personality traits; third, demonstrating PEFT's superiority over IKE in personality manipulation; and finally, analysing and validating emoji usage through explainability methods such as Mechanistic Interpretability and In-context learning Explainability methods.

cs.CL↗

arXiv2024

DeTEcT: Dynamic and Probabilistic Parameters Extension

This paper presents a theoretical extension of the DeTEcT framework proposed by Sadykhov et al., DeTEcT, where a formal analysis framework was introduced for modelling wealth distribution in token economies. DeTEcT is a framework for analysing economic activity, simulating macroeconomic scenarios, and algorithmically setting policies in token economies. This paper proposes four ways of parametrizing the framework, where dynamic vs static parametrization is considered along with the probabilistic vs non-probabilistic. Using these parametrization techniques, we demonstrate that by adding restrictions to the framework it is possible to derive the existing wealth distribution models from DeTEcT. In addition to exploring parametrization techniques, this paper studies how money supply in DeTEcT framework can be transformed to become dynamic, and how this change will affect the dynamics of wealth distribution. The motivation for studying dynamic money supply is that it enables DeTEcT to be applied to modelling token economies without maximum supply (i.e., Ethereum), and it adds constraints to the framework in the form of symmetries.

q-fin.GN↗

arXiv2024

THaMES: An End-to-End Tool for Hallucination Mitigation and Evaluation in Large Language Models

Hallucination, the generation of factually incorrect content, is a growing challenge in Large Language Models (LLMs). Existing detection and mitigation methods are often isolated and insufficient for domain-specific needs, lacking a standardized pipeline. This paper introduces THaMES (Tool for Hallucination Mitigations and EvaluationS), an integrated framework and library addressing this gap. THaMES offers an end-to-end solution for evaluating and mitigating hallucinations in LLMs, featuring automated test set generation, multifaceted benchmarking, and adaptable mitigation strategies. It automates test set creation from any corpus, ensuring high data quality, diversity, and cost-efficiency through techniques like batch processing, weighted sampling, and counterfactual validation. THaMES assesses a model's ability to detect and reduce hallucinations across various tasks, including text generation and binary classification, applying optimal mitigation strategies like In-Context Learning (ICL), Retrieval Augmented Generation (RAG), and Parameter-Efficient Fine-tuning (PEFT). Evaluations of state-of-the-art LLMs using a knowledge base of academic papers, political news, and Wikipedia reveal that commercial models like GPT-4o benefit more from RAG than ICL, while open-weight models like Llama-3.1-8B-Instruct and Mistral-Nemo gain more from ICL. Additionally, PEFT significantly enhances the performance of Llama-3.1-8B-Instruct in both evaluation tasks.

cs.CL↗

arXiv2024

HEARTS: A Holistic Framework for Explainable, Sustainable and Robust Text Stereotype Detection

Stereotypes are generalised assumptions about societal groups, and even state-of-the-art LLMs using in-context learning struggle to identify them accurately. Due to the subjective nature of stereotypes, where what constitutes a stereotype can vary widely depending on cultural, social, and individual perspectives, robust explainability is crucial. Explainable models ensure that these nuanced judgments can be understood and validated by human users, promoting trust and accountability. We address these challenges by introducing HEARTS (Holistic Framework for Explainable, Sustainable, and Robust Text Stereotype Detection), a framework that enhances model performance, minimises carbon footprint, and provides transparent, interpretable explanations. We establish the Expanded Multi-Grain Stereotype Dataset (EMGSD), comprising 57,201 labelled texts across six groups, including under-represented demographics like LGBTQ+ and regional stereotypes. Ablation studies confirm that BERT models fine-tuned on EMGSD outperform those trained on individual components. We then analyse a fine-tuned, carbon-efficient ALBERT-V2 model using SHAP to generate token-level importance values, ensuring alignment with human understanding, and calculate explainability confidence scores by comparing SHAP and LIME outputs...

cs.CL↗