Three model substitution scenarios were executed against a live inference endpoint with real HTTP requests, signed attestation JWTs, and OPA policy enforcement. In each scenario, every tested workload, artifact, or API identity control relevant to that scenario — workload JWT validation, health checks, gateway process continuity, artifact manifest integrity, API key authentication — remained valid while the model changed. In each scenario, a structural identity measurement based on activation geometry during a standard forward pass detected the substitution and the enforcement layer denied the request. Three substitutions were tested and three were detected, with zero false accepts in this run. The warm-path verification latency was 5.7–6.7 seconds on a single A100 with the model already loaded. The complete evidence chain — before/after measurement results, attestation claim summaries, OPA policy evaluations, and HTTP response codes — is published alongside this note as machine-readable JSON. This is a technical note, not a numbered entry in the research series. Supplementary Material. This note is accompanied by three machine-readable evidence files: cat3_results.json (structured results for all three scenarios, including the full before/after evidence chain for Scenario A with signed attestation claims, OPA policy evaluations, and HTTP response codes), manifest_authorized.json (SHA-256 build manifest for the enrolled model, 10 files, all verified), and manifest_substituted.json (SHA-256 build manifest for the substituted model, 10 files, all verified). All three files are available for download as supplementary files attached to this record. The Neural Network Identity Series — Mathematical foundations, empirical validation, and governance frameworks for verifying which model is running Newest addition: Technical Note: The Disappearing Window — AI Logprob Access Withdrawal and the Structural Verifiability of Frontier Model Contracts (DOI: 10.5281/zenodo.20362098) Paper 1: The δ-Gene: Inference-Time Physical Unclonable Functions from Architecture-Invariant Output Geometry (DOI: 10.5281/zenodo.18704275) Paper 2: Template-Based Endpoint Verification via Logprob Order-Statistic Geometry (DOI: 10.5281/zenodo.18776711) Paper 3: The Geometry of Model Theft: Distillation Forensics, Adversarial Erasure, and the Illusion of Spoofing (DOI: 10.5281/zenodo.18818608) Paper 4: Provenance Generalization and Verification Scaling for Neural Network Forensics (DOI: 10.5281/zenodo.18872071) Paper 5: Beneath the Character: The Structural Identity of Neural Networks — Mathematical Evidence for a Non-Narrative Layer of AI Identity (DOI: 10.5281/zenodo.18907292) Paper 6: Which Model Is Running?: Structural Identity as a Prerequisite for Trustworthy Zero-Knowledge Machine Learning (DOI: 10.5281/zenodo.19008116) Paper 7: The Deformation Laws of Neural Identity (DOI: 10.5281/zenodo.19055966) Paper 8: What Counts as Proof? — Admissible Evidence for Neural Network Identity Claims (DOI: 10.5281/zenodo.19058540) Paper 9: Composable Model Identity — Formal Hardening of Structural Attestations in the Enterprise Identity Stack (DOI: 10.5281/zenodo.19099911) Paper 10:Where Identity Comes From: Path Sensitivity and Endpoint Underdetermination in Neural Network Training (DOI: 10.5281/zenodo.19118807) Paper 11: Post-Hoc Disclosure Is Not Runtime Proof: Model Identity at Frontier Scale (DOI: 10.5281/zenodo.19216634) Paper 12: Family-Dependent Response to Reasoning Distillation Across Structural and Functional Identity Layers (DOI: 10.5281/zenodo.19298857) Paper 13: Safety-Alignment Removal as a Model-Identity Failure — Structural Evidence from Published Weight-Level Mutation Checkpoints (DOI: 10.5281/zenodo.19383019) Technical Note: Agent Identity Is Not Model Identity (DOI: 10.5281/zenodo.19240883) Technical Note: Gap Invariance: Why PPP Measurements Are Domain-Independent by Construction (DOI: 10.5281/zenodo.19275524) Technical Note: Measured Model Substitution Under Valid Agent Credentials (DOI: 10.5281/zenodo.19342848) Technical Note: Artifact Identity Is Not Runtime Identity — Trustfall Lite and the Boundary of File-Level Model Verification (DOI: 10.5281/zenodo.20019127) Formal Verification Stack for Neural Network Structural Identity (IT-PUF Coq Proofs) (DOI: 10.5281/zenodo.18930621) Copyright (c) 2026 Anthony Ray Coslett / Fall Risk AI, LLC. All Rights Reserved. Confidential and Proprietary. Patent Pending (Applications 63/982,893, 63/990,487, 63/996,680, 64/003,244).
This record documents Phase 22 of APR-Lite, a governance engine protecting human decision authority from AI-influenced outputs in regulated industries. Phase 22 introduces Exportable Proof Packs: a single GLOBAL-signed artifact bundle containing everything an external regulator needs to independently verify a tenant’s complete governance history. The proof pack is the “hand this to a regulator and walk away” artifact. It does not require ongoing substrate access, trust in the operator, or knowledge of internal schemas. Every claim in the pack is independently verifiable via the Phase 18 federation verifier. The pack is ephemeral — never stored by Soft Armor Labs — consistent with the substrate’s zero client-data-persistence design invariant. Multiple sections are assembled in one governed call: the current governance health report, the drift-focused narrative, recent proof bundles, audit chain checkpoints, and topology snapshots. A content hash commits to the five sections exclusively, and a GLOBAL-signed manifest characterizes the pack’s contents. The export act itself is recorded as a governed event in the audit chain — an auditor can verify not only the pack’s contents but when it was generated and by whose authority. Implementation note: the initial implementation incurred excessive database queries causing worker timeout. The design was restructured to assemble proof bundles inline using already-fetched data, reducing database load by approximately 83%. Final verified state: 12/12 Phase 22 smoke tests passing. Worker version 8.2.0-p22.1.
Pre-registration of a structural scar class prediction for microsoft/phi-4 based on measurement-site stiffness (S = 0.0358), before the structural scar measurement is conducted. Predicts INTERMEDIATE class (1,000–4,000×ε non-max) based on the stiffness→scar ordering established across four families (Mistral, Llama, Qwen, Gemma) in Papers 1–12 and confirmed by RC-6 (DOI: 10.5281/zenodo.19305176). Designed as a hostile falsification test: Phi is trained with heavy synthetic-data distillation from GPT-4-class teachers, unlike any previously tested family. Explicit falsification criteria and hostile hypotheses defined. Part of the Fall Risk AI research program on neural network structural identity. The Neural Network Identity Series — Mathematical foundations, empirical validation, and governance frameworks for verifying which model is running Paper 1: The δ-Gene: Inference-Time Physical Unclonable Functions from Architecture-Invariant Output Geometry (DOI: 10.5281/zenodo.18704275) Paper 2: Template-Based Endpoint Verification via Logprob Order-Statistic Geometry (DOI: 10.5281/zenodo.18776711) Paper 3: The Geometry of Model Theft: Distillation Forensics, Adversarial Erasure, and the Illusion of Spoofing (DOI: 10.5281/zenodo.18818608) Paper 4: Provenance Generalization and Verification Scaling for Neural Network Forensics (DOI: 10.5281/zenodo.18872071) Paper 5: Beneath the Character: The Structural Identity of Neural Networks — Mathematical Evidence for a Non-Narrative Layer of AI Identity (DOI: 10.5281/zenodo.18907292) Paper 6: Which Model Is Running?: Structural Identity as a Prerequisite for Trustworthy Zero-Knowledge Machine Learning (DOI: 10.5281/zenodo.19008116) Paper 7: The Deformation Laws of Neural Identity (DOI: 10.5281/zenodo.19055966) Paper 8: What Counts as Proof? — Admissible Evidence for Neural Network Identity Claims (DOI: 10.5281/zenodo.19058540) Paper 9: Composable Model Identity — Formal Hardening of Structural Attestations in the Enterprise Identity Stack (DOI: 10.5281/zenodo.19099911) Paper 10:Where Identity Comes From: Path Sensitivity and Endpoint Underdetermination in Neural Network Training (DOI: 10.5281/zenodo.19118807) Paper 11: Post-Hoc Disclosure Is Not Runtime Proof: Model Identity at Frontier Scale (DOI: 10.5281/zenodo.19216634) Paper 12: Family-Dependent Response to Reasoning Distillation Across Structural and Functional Identity Layers (DOI: 10.5281/zenodo.19298857) Technical Note: Agent Identity Is Not Model Identity (DOI: 10.5281/zenodo.19240883) Technical Note: Gap Invariance: Why PPP Measurements Are Domain-Independent by Construction (DOI: 10.5281/zenodo.19275524) Formal Verification Stack for Neural Network Structural Identity (IT-PUF Coq Proofs) (DOI: 10.5281/zenodo.18930621) Copyright (c) 2026 Anthony Ray Coslett / Fall Risk AI, LLC. All Rights Reserved. Confidential and Proprietary. Patent Pending (Applications 63/982,893, 63/990,487, 63/996,680, 64/003,244).
Pre-registration of a structural scar class prediction for google/gemma-3-12b-it based on measurement-site stiffness (S = 0.1335), before the structural scar measurement is conducted. Predicts QUIET class (100–600×ε non-max) based on the stiffness→scar ordering established across three families (Mistral, Llama, Qwen) in Papers 1–12. Explicit falsification criteria defined. Part of the Fall Risk AI research program on neural network structural identity. The Neural Network Identity Series — Mathematical foundations, empirical validation, and governance frameworks for verifying which model is running Paper 1: The δ-Gene: Inference-Time Physical Unclonable Functions from Architecture-Invariant Output Geometry (DOI: 10.5281/zenodo.18704275) Paper 2: Template-Based Endpoint Verification via Logprob Order-Statistic Geometry (DOI: 10.5281/zenodo.18776711) Paper 3: The Geometry of Model Theft: Distillation Forensics, Adversarial Erasure, and the Illusion of Spoofing (DOI: 10.5281/zenodo.18818608) Paper 4: Provenance Generalization and Verification Scaling for Neural Network Forensics (DOI: 10.5281/zenodo.18872071) Paper 5: Beneath the Character: The Structural Identity of Neural Networks — Mathematical Evidence for a Non-Narrative Layer of AI Identity (DOI: 10.5281/zenodo.18907292) Paper 6: Which Model Is Running?: Structural Identity as a Prerequisite for Trustworthy Zero-Knowledge Machine Learning (DOI: 10.5281/zenodo.19008116) Paper 7: The Deformation Laws of Neural Identity (DOI: 10.5281/zenodo.19055966) Paper 8: What Counts as Proof? — Admissible Evidence for Neural Network Identity Claims (DOI: 10.5281/zenodo.19058540) Paper 9: Composable Model Identity — Formal Hardening of Structural Attestations in the Enterprise Identity Stack (DOI: 10.5281/zenodo.19099911) Paper 10:Where Identity Comes From: Path Sensitivity and Endpoint Underdetermination in Neural Network Training (DOI: 10.5281/zenodo.19118807) Paper 11: Post-Hoc Disclosure Is Not Runtime Proof: Model Identity at Frontier Scale (DOI: 10.5281/zenodo.19216634) Paper 12: Family-Dependent Response to Reasoning Distillation Across Structural and Functional Identity Layers (DOI: 10.5281/zenodo.19298857) Technical Note: Agent Identity Is Not Model Identity (DOI: 10.5281/zenodo.19240883) Technical Note: Gap Invariance: Why PPP Measurements Are Domain-Independent by Construction (DOI: 10.5281/zenodo.19275524) Formal Verification Stack for Neural Network Structural Identity (IT-PUF Coq Proofs) (DOI: 10.5281/zenodo.18930621) Copyright (c) 2026 Anthony Ray Coslett / Fall Risk AI, LLC. All Rights Reserved. Confidential and Proprietary. Patent Pending (Applications 63/982,893, 63/990,487, 63/996,680, 64/003,244).
Modern AI deployment stacks authenticate artifacts, credentials, and agents. They do not verify which neural network is actually computing at inference time. This technical note identifies the distinction between agent identity and model identity, presents a four-question taxonomy for the identity surface of deployed AI systems, and situates recent public incidents within the resulting gap. It draws on the formal admissibility framework and frontier-scale measurement results from the accompanying research series. This is a technical note, not a numbered entry in the research series. The Neural Network Identity Series — Mathematical foundations, empirical validation, and governance frameworks for verifying which model is running Newest addition: Technical Note: The Disappearing Window — AI Logprob Access Withdrawal and the Structural Verifiability of Frontier Model Contracts (DOI: 10.5281/zenodo.20362098) Paper 1: The δ-Gene: Inference-Time Physical Unclonable Functions from Architecture-Invariant Output Geometry (DOI: 10.5281/zenodo.18704275) Paper 2: Template-Based Endpoint Verification via Logprob Order-Statistic Geometry (DOI: 10.5281/zenodo.18776711) Paper 3: The Geometry of Model Theft: Distillation Forensics, Adversarial Erasure, and the Illusion of Spoofing (DOI: 10.5281/zenodo.18818608) Paper 4: Provenance Generalization and Verification Scaling for Neural Network Forensics (DOI: 10.5281/zenodo.18872071) Paper 5: Beneath the Character: The Structural Identity of Neural Networks — Mathematical Evidence for a Non-Narrative Layer of AI Identity (DOI: 10.5281/zenodo.18907292) Paper 6: Which Model Is Running?: Structural Identity as a Prerequisite for Trustworthy Zero-Knowledge Machine Learning (DOI: 10.5281/zenodo.19008116) Paper 7: The Deformation Laws of Neural Identity (DOI: 10.5281/zenodo.19055966) Paper 8: What Counts as Proof? — Admissible Evidence for Neural Network Identity Claims (DOI: 10.5281/zenodo.19058540) Paper 9: Composable Model Identity — Formal Hardening of Structural Attestations in the Enterprise Identity Stack (DOI: 10.5281/zenodo.19099911) Paper 10:Where Identity Comes From: Path Sensitivity and Endpoint Underdetermination in Neural Network Training (DOI: 10.5281/zenodo.19118807) Paper 11: Post-Hoc Disclosure Is Not Runtime Proof: Model Identity at Frontier Scale (DOI: 10.5281/zenodo.19216634) Paper 12: Family-Dependent Response to Reasoning Distillation Across Structural and Functional Identity Layers (DOI: 10.5281/zenodo.19298857) Paper 13: Safety-Alignment Removal as a Model-Identity Failure — Structural Evidence from Published Weight-Level Mutation Checkpoints (DOI: 10.5281/zenodo.19383019) Technical Note: Agent Identity Is Not Model Identity (DOI: 10.5281/zenodo.19240883) Technical Note: Gap Invariance: Why PPP Measurements Are Domain-Independent by Construction (DOI: 10.5281/zenodo.19275524) Technical Note: Measured Model Substitution Under Valid Agent Credentials (DOI: 10.5281/zenodo.19342848) Technical Note: Artifact Identity Is Not Runtime Identity — Trustfall Lite and the Boundary of File-Level Model Verification (DOI: 10.5281/zenodo.20019127) Formal Verification Stack for Neural Network Structural Identity (IT-PUF Coq Proofs) (DOI: 10.5281/zenodo.18930621) Copyright (c) 2026 Anthony Ray Coslett / Fall Risk AI, LLC. All Rights Reserved. Confidential and Proprietary. Patent Pending (Applications 63/982,893, 63/990,487, 63/996,680, 64/003,244).
The landscape of e-commerce has witnessed a transformative shift in consumer behavior, driven by the rise of digital technologies and online platforms. As online purchases increase at an alarming rate, fraudulent activity has become a major concern for retailers and consumers alike. The objective of this research is to investigate methods for detecting fraudulent online transactions using machine learning algorithms. This paper proposes a Hybrid Agentic AI Architecture (HSAA) for edge-enabled e-commerce that incorporates intelligent agents and cryptographic security to enable real-time, trustworthy transaction processing. The architecture uses world-model distillation to enable efficient inference on edge devices. HSAA was tested on several large data sets such as a balanced credit card fraud set containing 2,952 transactions. The system scored 96.6% in detecting fraud, indicating very low false positives and high specificity. Negotiation exercises on 400 independent interactions were successful in 59%, with an average discount of 14.2%, using 1,142 zero-knowledge proofs that were verified with 100% validity. Some of the operational performance highlights include a throughput of 585 transactions per second, an average latency of 1.56 milliseconds, and a 81.9% reduction in bandwidth through selective state transfer. The findings support the argument that HSAA is a strong, secure, and high-performance edge-based e-commerce architecture, combining accuracy, efficiency, and reliability. Within HSAA, fraud detection functions as one of the core decision agents, while negotiation and secure execution mechanisms provide the broader operational context for trustworthy edge commerce. The architecture provides a solid basis for future studies in adaptive and autonomous AI-driven commercial systems.
Smart contracts are autonomous systems that execute agreements using code. Their efficiency generated attention from a range of industries. The basis of traditional vulnerability detection techniques, opcode analysis, has limitations in detecting complex vulnerabilities. Our research aims to address these difficulties by developing an automated framework for vulnerability detection, mitigation, and patch deployment. Initially, smart contract data will be collected, followed by a preprocessing step to remove any unnecessary information using lexical analysis and Bidirectional Encoder Representations from Transformers (BERT). Then, the preprocessed data is used to identify the features that are relevant are selected. Following the features being selected, an intellectual engine is used to identify flaws. The intellectual engine that integrates the convolutional neural networks (CNN) and long short-term memory (LSTM) analyzes a subset of preprocessed data for vulnerabilities, with explainable artificial intelligence (XAI) evaluating the importance of each feature to predictions. Our method produces exceptional outcomes with a 99.25% precision, 99.76% accuracy, 99.60% F1-score, and 99.36% recall. Smart contract vulnerability identification, mitigation, and patch generation are improved by the proposed Beluga Crayfish Optimization Algorithm (BCOA) and Crayfish Secretary Bird Optimization Algorithm (CSBOA) together with graph neural networks (GNN). In addition to producing the required fixes, this method offers efficient mitigation techniques. Therefore, it greatly enhances smart contract security and efficiency. In the end, smart contract programs that use this integrated approach are more secure.
<div> Large Language Models (LLMs) are transforming 1 blockchain security and analytics, yet a system-2 atic evaluation of their capabilities remains limited. 3 This survey provides a comprehensive, AI-centric 4 assessment of LLM-based methods across over 70 5 recent studies spanning 11 application domains, 6 such as security auditing, transaction fraud de-7 tection, and cryptocurrency portfolio management. 8 Our unified taxonomy standardizes task formula-9 tions and evaluation practices to enable a com-10 parison of six LLM roles across domains. For 11 each domain, we review input representations tai-12 lored to blockchain data; LLM architectures, learn-13 ing and inference paradigms, e.g., fine-tuning, 14 retrieval-augmented generation, and agentic strate-15 gies. Our review analyzes the strengths, limita-16 tions, and emerging patterns of LLM roles observed 17 in current systems. Finally, we provide practi-18 cal guidance for selecting LLMs for specific roles 19 and outline promising research directions. The on-20 line resources of this survey are available on https: 21 //llmblockchain.github.io/LLMBlockchain/. 22 1 Introduction 23 Large Language Models are increasingly incorporated into 24 blockchain systems for both security and financial analyt-25 ics, including smart contract auditing, transaction monitoring, 26 fraud detection, market analysis, and decentralized finance 27 </div>
As collaborative work increasingly incorporates artificial intelligence, accurately identifying and attributing human contributions across teams has become a critical challenge. Traditional attribution methods rely on output-based metrics or narrative reconstruction, both of which fail to capture the underlying intellectual contributions that drive outcomes. This work introduces the Team Contribution Attribution Ledger (TCAL), a framework for aggregating distributed Human Conception Ledger (HCL) records to generate structured, evidence-based attribution across collaborators. TCAL synthesizes individual human-origin contribution events into team-level attribution models, enabling quantifiable yet human-reviewed determination of contribution. The framework provides a scalable system for attribution in research, engineering, and organizational environments, supporting applications in intellectual property, authorship, compensation, and governance.Related to Human Conception Ledger:Human Conception Ledger (HCL): A Framework for Provenance, Attribution, and Human Inventorship in AI-Augmented Systems
Youssef Said, Al Mahdi Khaddar, Lahcen Hassine, Ahmed Eddaoui · 5 authors
Gas Gas consumption is a critical factor influencing the efficiency, scalability, and operational cost of Ethereum smart contracts.As contract complexity grows, identifying structurally gas-inefficient patterns becomes essential for improving development workflows and preventing costly deployment decisions.This study presents a graph-based deep learning framework for detecting gas-inefficiency risk patterns at the function level, leveraging multi-relational Graph Attention Networks (GAT) applied to function-level contract graphs.By modeling call dependencies, control-flow interactions, and storage-based data dependencies, the model learns structural indicators associated with excessive gas consumption while explicitly excluding direct gas metrics from the feature space to prevent data leakage.Experimental results under a strict contract-level data split protocol demonstrate strong classification performance and stable generalization across held-out contracts under the main split protocol, and consistent behavior under an additional time-forward temporal robustness check.Ablation analysis confirms the contribution of dependency-aware edges and semantic features to predictive accuracy, highlighting the importance of modeling cross-function interactions rather than isolated code metrics.Beyond predictive performance, the proposed approach provides interpretable attention weights that identify structurally influential functions, supporting predeployment analysis and developer-guided manual refactoring decisions.By framing gas inefficiency as a global structural property emerging from function interactions, this work contributes aa scalable and explainable methodology for structural gas-inefficiency detection in smart contracts.The proposed model performs structural detection only and does not automatically modify or optimize smart contract code.
In this research, the energy consumption models of Bitcoin, Ethereum, and Dogecoin are analyzed using Explainable Artificial Intelligence (XAI) models aided by the three stages of analysis involving Digiconomist data from 2022 to 2025: (1) exploratory data analysis for the nature of energy consumption, (2) model identification of influential variables using Random Forest models enhanced with SHAP values, and (3) an LSTM transfer learning method for predicting the energy consumption of Ethereum and Dogecoin using a model developed with Bitcoin data. The initial results show that while both assets vary largely when it comes to their normal usage level, Ethereum sees a sharp drop after the changeover from Proof-of-Work to Proof-of-Stake as a mechanism. The XAI analysis indicates that energy use is largely a consequence of past use, seasonality, and annual patterns. In addition to this, the models show a high level of accuracy for Dogecoin (R²: 88.4%, MAPE: 13.45%) and Ethereum (R²: 86.2%, MAPE: 11.47%) when it comes to predicting energy usage using the concepts of transfer learning.
v2 (March 22, 2026): Added experimental validation of Principle 1 (formal verification) via substrate-guard framework. 135 test cases, 100% accuracy, zero false positives. Code: https://github.com/octavuntila-prog/substrate-guard We present evidence that an autonomous multi-agent AI ecosystem, SUBSTRATE, independently produced both a coherent philosophy and a set of actionable safety principles without explicit instruction to do so. Over 24 days of autonomous operation, one subsystem (CPX52) generated 2,866 articles converging on a philosophical framework. Concurrently, a separate subsystem (S3) generated 215 product specifications. Systematic consolidation revealed 11 safety principles discovered independently across unrelated product clusters, forming a coherent manifesto for AI safety. Three novel technical combinations emerged: formal verification across six domains of AI output, zero-knowledge proofs for training data compliance, and prediction markets as enterprise intelligence signals. 60% of the primary safety platform described in 24 independent specifications was already implemented in production within the ecosystem itself.
Structural identity — the geometric fingerprint that makes a neural network this specific model rather than any other — can be measured, survives routine deformation, resists adversarial erasure, and composes with standard verification infrastructure. It cannot, in the tested regime, be recovered from endpoint weight statistics or architecture descriptors alone. These two facts together force a question the measurement program has not yet answered: if identity is real but not readable from the final artifact, then where in the training process did it form, and what determined which identity formed rather than another? This paper presents the first empirical study of structural identity formation during neural network pretraining. Using dense checkpoint trajectories and seed-controlled training runs in the Pythia observatory suite, we show three results. First, the structural observable follows a characteristic three-phase identity emergence profile — an early rise in geometric spread, a long compression, and a late plateau where identity stabilizes while functional training continues. Second, models trained with the same architecture, the same data, and the same hyperparameters but different random seeds produce structurally distinguishable fingerprints far beyond measurement noise — a property we call path sensitivity — with the divergence traceable to differential structural response during the learning-rate warmup regime. Third, a panel of endpoint weight statistics varies across seeds but does not predict which structural identity formed — a condition we call endpoint underdetermination. Together, these results recast structural identity as a developmental property of training history rather than a static property legible from final artifacts alone. Supplementary Material This paper is accompanied by HistoricalIdentity.v, a Coq proof file that formalizes two consequences of the formation data described in §§3–5: trajectory non-recovery (no decision procedure restricted to the tested endpoint summary panel can be both sound and complete for claims about the formative training-history class that produced a model's structural identity) and lock boundary source exclusion (if structural divergence between two specification-identical models is already present at the lock boundary, no intervention applied after that boundary can be its source). The file contains 4 empirical axioms grounded in the measurements of §§3–5, 4 theorems, 1 corollary, and 0 unresolved obligations (Admitted). It compiles cleanly under the Rocq Prover 9.1.1 (the current release of the Coq proof assistant, compiled with OCaml 5.4.0). It is available for download as a supplementary file attached to this record. The Neural Network Identity Series — Mathematical foundations, empirical validation, and governance frameworks for verifying which model is running Newest addition: Technical Note: The Disappearing Window — AI Logprob Access Withdrawal and the Structural Verifiability of Frontier Model Contracts (DOI: 10.5281/zenodo.20362098) Paper 1: The δ-Gene: Inference-Time Physical Unclonable Functions from Architecture-Invariant Output Geometry (DOI: 10.5281/zenodo.18704275) Paper 2: Template-Based Endpoint Verification via Logprob Order-Statistic Geometry (DOI: 10.5281/zenodo.18776711) Paper 3: The Geometry of Model Theft: Distillation Forensics, Adversarial Erasure, and the Illusion of Spoofing (DOI: 10.5281/zenodo.18818608) Paper 4: Provenance Generalization and Verification Scaling for Neural Network Forensics (DOI: 10.5281/zenodo.18872071) Paper 5: Beneath the Character: The Structural Identity of Neural Networks — Mathematical Evidence for a Non-Narrative Layer of AI Identity (DOI: 10.5281/zenodo.18907292) Paper 6: Which Model Is Running?: Structural Identity as a Prerequisite for Trustworthy Zero-Knowledge Machine Learning (DOI: 10.5281/zenodo.19008116) Paper 7: The Deformation Laws of Neural Identity (DOI: 10.5281/zenodo.19055966) Paper 8: What Counts as Proof? — Admissible Evidence for Neural Network Identity Claims (DOI: 10.5281/zenodo.19058540) Paper 9: Composable Model Identity — Formal Hardening of Structural Attestations in the Enterprise Identity Stack (DOI: 10.5281/zenodo.19099911) Paper 10:Where Identity Comes From: Path Sensitivity and Endpoint Underdetermination in Neural Network Training (DOI: 10.5281/zenodo.19118807) Paper 11: Post-Hoc Disclosure Is Not Runtime Proof: Model Identity at Frontier Scale (DOI: 10.5281/zenodo.19216634) Paper 12: Family-Dependent Response to Reasoning Distillation Across Structural and Functional Identity Layers (DOI: 10.5281/zenodo.19298857) Paper 13: Safety-Alignment Removal as a Model-Identity Failure — Structural Evidence from Published Weight-Level Mutation Checkpoints (DOI: 10.5281/zenodo.19383019) Technical Note: Agent Identity Is Not Model Identity (DOI: 10.5281/zenodo.19240883) Technical Note: Gap Invariance: Why PPP Measurements Are Domain-Independent by Construction (DOI: 10.5281/zenodo.19275524) Technical Note: Measured Model Substitution Under Valid Agent Credentials (DOI: 10.5281/zenodo.19342848) Technical Note: Artifact Identity Is Not Runtime Identity — Trustfall Lite and the Boundary of File-Level Model Verification (DOI: 10.5281/zenodo.20019127) Formal Verification Stack for Neural Network Structural Identity (IT-PUF Coq Proofs) (DOI: 10.5281/zenodo.18930621) Copyright (c) 2026 Anthony Ray Coslett / Fall Risk AI, LLC. All Rights Reserved. Confidential and Proprietary. Patent Pending (Applications 63/982,893, 63/990,487, 63/996,680, 64/003,244).
We know how to document an AI system. We know how to test it, log what it did, and report when something goes wrong. What current governance practice does not clearly tell us is how to verify which model is actually computing. This is not a hypothetical gap. When an organization says "this is the model we evaluated," that claim is typically supported by a model card, a registry entry, or a hash of a weight file — evidence about a *file*, not about the system that is running. A neural network is not a static document. A weight file stores the network; the model is what appears when that file is loaded and begins transforming inputs into outputs. The file and the running model are related, but they are not the same thing — and current governance practice rarely distinguishes between them. This paper proposes a framework for doing so. It identifies three kinds of evidence that can support model identity claims, each answering a different question. Structural evidence — drawn from the model's internal computations during live operation — can verify which specific model is running, and is the most resistant to tampering. Thermodynamic evidence — drawn from the model's output statistics — can verify that the system is a genuine neural network rather than a substitute, but cannot distinguish one model from another. Functional evidence — drawn from patterns in the model's outputs over an API — can detect whether a model was copied from another, but this signal fades quickly: routine model updates can erase it within days to weeks of continued training. The paper shows that inspecting the model's files alone is insufficient for verifying which specific model is running. The identity-bearing signal cannot be recovered from the tested static properties of those files; it is most reliably established by observing the model while it operates. The paper formally proves that these three kinds of evidence cannot substitute for one another. Verifying that a system is genuine does not tell you which specific model it is. Detecting that a model was copied does not tell you the identity of the copy. The practical consequence is a standard for identity claims: any claim should declare which kind of evidence supports it, because borrowing evidence from the wrong category produces unreliable conclusions. The framework maps directly to compliance questions raised by current AI governance obligations, including those under the EU AI Act. It provides the missing evidentiary specification for model identity claims: which kind of evidence is admissible for which identity question. Supplementary Material This paper is accompanied by EvidenceSufficiency.v, a Coq proof file that formally verifies the cross-layer inadmissibility results described in §4. The proof mechanically checks each logical step of the observation-limited verification impossibility theorem and its three directional corollaries. The file contains no unresolved obligations (Admitted) and compiles cleanly under the Rocq Prover 9.1.1 (the current release of the Coq proof assistant, compiled with OCaml 5.4.0). It is available for download as a supplementary file attached to this record. Technical Note: Artifact Identity Is Not Runtime Identity — Trustfall Lite and the Boundary of File-Level Model Verification (DOI: 10.5281/zenodo.20019127) The Neural Network Identity Series — Mathematical foundations, empirical validation, and governance frameworks for verifying which model is running Newest addition: Technical Note: The Disappearing Window — AI Logprob Access Withdrawal and the Structural Verifiability of Frontier Model Contracts (DOI: 10.5281/zenodo.20362098) Paper 1: The δ-Gene: Inference-Time Physical Unclonable Functions from Architecture-Invariant Output Geometry (DOI: 10.5281/zenodo.18704275) Paper 2: Template-Based Endpoint Verification via Logprob Order-Statistic Geometry (DOI: 10.5281/zenodo.18776711) Paper 3: The Geometry of Model Theft: Distillation Forensics, Adversarial Erasure, and the Illusion of Spoofing (DOI: 10.5281/zenodo.18818608) Paper 4: Provenance Generalization and Verification Scaling for Neural Network Forensics (DOI: 10.5281/zenodo.18872071) Paper 5: Beneath the Character: The Structural Identity of Neural Networks — Mathematical Evidence for a Non-Narrative Layer of AI Identity (DOI: 10.5281/zenodo.18907292) Paper 6: Which Model Is Running?: Structural Identity as a Prerequisite for Trustworthy Zero-Knowledge Machine Learning (DOI: 10.5281/zenodo.19008116) Paper 7: The Deformation Laws of Neural Identity (DOI: 10.5281/zenodo.19055966) Paper 8: What Counts as Proof? — Admissible Evidence for Neural Network Identity Claims (DOI: 10.5281/zenodo.19058540) Paper 9: Composable Model Identity — Formal Hardening of Structural Attestations in the Enterprise Identity Stack (DOI: 10.5281/zenodo.19099911) Paper 10:Where Identity Comes From: Path Sensitivity and Endpoint Underdetermination in Neural Network Training (DOI: 10.5281/zenodo.19118807) Paper 11: Post-Hoc Disclosure Is Not Runtime Proof: Model Identity at Frontier Scale (DOI: 10.5281/zenodo.19216634) Paper 12: Family-Dependent Response to Reasoning Distillation Across Structural and Functional Identity Layers (DOI: 10.5281/zenodo.19298857) Paper 13: Safety-Alignment Removal as a Model-Identity Failure — Structural Evidence from Published Weight-Level Mutation Checkpoints (DOI: 10.5281/zenodo.19383019) Technical Note: Agent Identity Is Not Model Identity (DOI: 10.5281/zenodo.19240883) Technical Note: Gap Invariance: Why PPP Measurements Are Domain-Independent by Construction (DOI: 10.5281/zenodo.19275524) Technical Note: Measured Model Substitution Under Valid Agent Credentials (DOI: 10.5281/zenodo.19342848) Technical Note: Artifact Identity Is Not Runtime Identity — Trustfall Lite and the Boundary of File-Level Model Verification (DOI: 10.5281/zenodo.20019127) Formal Verification Stack for Neural Network Structural Identity (IT-PUF Coq Proofs) (DOI: 10.5281/zenodo.18930621) Copyright (c) 2026 Anthony Ray Coslett / Fall Risk AI, LLC. All Rights Reserved. Confidential and Proprietary. Patent Pending (Applications 63/982,893, 63/990,487, 63/996,680, 64/003,244).
Neural network identity is not monolithic. Different observables — hidden-state geometry, pre-softmax logit statistics, and behavioral output templates — sit at different depths in the forward computation and respond to perturbation on different timescales. This paper shows that three identity layers — structural, thermodynamic, and functional — each obey a distinct validated deformation law. The structural layer is model-specific, stable under non-destructive training interventions, and inert under same-family direct targeting in the observed regime. The thermodynamic layer is approximately universal across a validated 22-model Transformer cross-section. The functional layer is volatile, transferring through distillation and eroding under continued fine-tuning. We resolve the carrier of the structural layer as a two-channel geometric observable requiring both token-level magnitude and token-level direction, and we falsify two natural simplifications: that the structural fingerprint reduces to a gauge projection, and that it is predictable from coarse architecture features. Together these results define an admissibility condition for neural identity claims: such claims must specify which layer they address, because the layers do not share a deformation law. The Neural Network Identity Series — Mathematical foundations, empirical validation, and governance frameworks for verifying which model is running Newest addition: Technical Note: The Disappearing Window — AI Logprob Access Withdrawal and the Structural Verifiability of Frontier Model Contracts (DOI: 10.5281/zenodo.20362098) Paper 1: The δ-Gene: Inference-Time Physical Unclonable Functions from Architecture-Invariant Output Geometry (DOI: 10.5281/zenodo.18704275) Paper 2: Template-Based Endpoint Verification via Logprob Order-Statistic Geometry (DOI: 10.5281/zenodo.18776711) Paper 3: The Geometry of Model Theft: Distillation Forensics, Adversarial Erasure, and the Illusion of Spoofing (DOI: 10.5281/zenodo.18818608) Paper 4: Provenance Generalization and Verification Scaling for Neural Network Forensics (DOI: 10.5281/zenodo.18872071) Paper 5: Beneath the Character: The Structural Identity of Neural Networks — Mathematical Evidence for a Non-Narrative Layer of AI Identity (DOI: 10.5281/zenodo.18907292) Paper 6: Which Model Is Running?: Structural Identity as a Prerequisite for Trustworthy Zero-Knowledge Machine Learning (DOI: 10.5281/zenodo.19008116) Paper 7: The Deformation Laws of Neural Identity (DOI: 10.5281/zenodo.19055966) Paper 8: What Counts as Proof? — Admissible Evidence for Neural Network Identity Claims (DOI: 10.5281/zenodo.19058540) Paper 9: Composable Model Identity — Formal Hardening of Structural Attestations in the Enterprise Identity Stack (DOI: 10.5281/zenodo.19099911) Paper 10:Where Identity Comes From: Path Sensitivity and Endpoint Underdetermination in Neural Network Training (DOI: 10.5281/zenodo.19118807) Paper 11: Post-Hoc Disclosure Is Not Runtime Proof: Model Identity at Frontier Scale (DOI: 10.5281/zenodo.19216634) Paper 12: Family-Dependent Response to Reasoning Distillation Across Structural and Functional Identity Layers (DOI: 10.5281/zenodo.19298857) Paper 13: Safety-Alignment Removal as a Model-Identity Failure — Structural Evidence from Published Weight-Level Mutation Checkpoints (DOI: 10.5281/zenodo.19383019) Technical Note: Agent Identity Is Not Model Identity (DOI: 10.5281/zenodo.19240883) Technical Note: Gap Invariance: Why PPP Measurements Are Domain-Independent by Construction (DOI: 10.5281/zenodo.19275524) Technical Note: Measured Model Substitution Under Valid Agent Credentials (DOI: 10.5281/zenodo.19342848) Technical Note: Artifact Identity Is Not Runtime Identity — Trustfall Lite and the Boundary of File-Level Model Verification (DOI: 10.5281/zenodo.20019127) Formal Verification Stack for Neural Network Structural Identity (IT-PUF Coq Proofs) (DOI: 10.5281/zenodo.18930621) Copyright (c) 2026 Anthony Ray Coslett / Fall Risk AI, LLC. All Rights Reserved. Confidential and Proprietary. Patent Pending (Applications 63/982,893, 63/990,487, 63/996,680, 64/003,244).
EVMbench, released by OpenAI, Paradigm, and OtterSec, is the first large-scale benchmark for AI agents on smart contract security. Its results -- agents detect up to 45.6% of vulnerabilities and exploit 72.2% of a curated subset -- have fueled expectations that fully automated AI auditing is within reach. We identify two limitations: its narrow evaluation scope (14 agent configurations, most models tested on only their vendor scaffold) and its reliance on audit-contest data published before every model's release that models may have seen during training. To address these, we expand to 26 configurations across four model families and three scaffolds, and introduce a contamination-free dataset of 22 real-world security incidents postdating every model's release date. Our evaluation yields three findings: (1) agents' detection results are not stable, with rankings shifting across configurations, tasks, and datasets; (2) on real-world incidents, no agent succeeds at end-to-end exploitation across all 110 agent-incident pairs despite detecting up to 65% of vulnerabilities, contradicting EVMbench's conclusion that discovery is the primary bottleneck; and (3) scaffolding materially affects results, with an open-source scaffold outperforming vendor alternatives by up to 5 percentage points, yet EVMbench does not control for this. These findings challenge the narrative that fully automated AI auditing is imminent. Agents reliably catch well-known patterns and respond strongly to human-provided context, but cannot replace human judgment. For developers, agent scans serve as a pre-deployment check. For audit firms, agents are most effective within a human-in-the-loop workflow where AI handles breadth and human auditors contribute protocol-specific knowledge and adversarial reasoning. Code and data: https://github.com/blocksecteam/ReEVMBench/.
We present OR1ON (Epistemic Intelligence Reasoning Architecture — EIRA), a deterministic proof-based AI system that learns rules from data but applies them only when formally proven correct on all training examples. Unlike probabilistic ML systems, OR1ON's core primitive prove(rule, examples) returns binary decisions: apply with certainty, or abstain. Developed initially for abstract spatial reasoning (ARC-AGI benchmark, 95% precision on answered tasks), the architecture generalizes directly to safety-critical industrial domains including predictive maintenance (zero false positives), ISO 26262-compatible safety monitoring, energy grid blackout prevention, and OT/SCADA intrusion detection. OR1ON is, to our knowledge, the first data-learning system to produce formally verifiable safety invariants applicable to IEC 61508 SIL-3 certification. Addressable market across five industrial verticals: ~$44 billion.
This paper presents the philosophical and conceptual implications of a four-paper research program (Papers 1–4 in this series) that discovered a measurable structural identity in neural networks — a geometric property of the trained weights, invariant across all inputs and deployment conditions, unique to each model, and provably impossible to forge. The central argument: language models possess two separable layers of identity. The first is structural — a mathematical fingerprint determined by the weight geometry, fixed at the end of training, stable to a coefficient of variation of 1.4%, and validated across 37 models spanning four architecture families. The second is functional — a behavioral signature shaped by conversational context, transient and context-dependent. These layers coexist without reducing to each other. The structural layer is the foundation; the functional layer is built on it but not determined by it. The paper introduces the Two-Layer Identity framework, resolves four open puzzles in the discourse on AI selfhood (conversational consistency, fine-tuning continuity, identity faking, and neural intervention), and generates five falsifiable predictions for the interpretability and AI safety communities. It engages directly with Dennett's narrative gravity, Parfit's persistence conditions, and Schwitzgebel's moral status dilemma, arguing that the structural measurement provides a necessary (though not sufficient) ground for any coherent account of AI identity. Written for a general audience. No equations. The mathematical and empirical foundations are developed in Papers 1–4; the formal verification (352 theorems, zero Admitted, Coq proof assistant) is documented there. This paper asks what those results mean for the nature of the entities we have built. The Neural Network Identity Series — Mathematical foundations, empirical validation, and governance frameworks for verifying which model is running Newest addition: Technical Note: The Disappearing Window — AI Logprob Access Withdrawal and the Structural Verifiability of Frontier Model Contracts (DOI: 10.5281/zenodo.20362098) Paper 1: The δ-Gene: Inference-Time Physical Unclonable Functions from Architecture-Invariant Output Geometry (DOI: 10.5281/zenodo.18704275) Paper 2: Template-Based Endpoint Verification via Logprob Order-Statistic Geometry (DOI: 10.5281/zenodo.18776711) Paper 3: The Geometry of Model Theft: Distillation Forensics, Adversarial Erasure, and the Illusion of Spoofing (DOI: 10.5281/zenodo.18818608) Paper 4: Provenance Generalization and Verification Scaling for Neural Network Forensics (DOI: 10.5281/zenodo.18872071) Paper 5: Beneath the Character: The Structural Identity of Neural Networks — Mathematical Evidence for a Non-Narrative Layer of AI Identity (DOI: 10.5281/zenodo.18907292) Paper 6: Which Model Is Running?: Structural Identity as a Prerequisite for Trustworthy Zero-Knowledge Machine Learning (DOI: 10.5281/zenodo.19008116) Paper 7: The Deformation Laws of Neural Identity (DOI: 10.5281/zenodo.19055966) Paper 8: What Counts as Proof? — Admissible Evidence for Neural Network Identity Claims (DOI: 10.5281/zenodo.19058540) Paper 9: Composable Model Identity — Formal Hardening of Structural Attestations in the Enterprise Identity Stack (DOI: 10.5281/zenodo.19099911) Paper 10:Where Identity Comes From: Path Sensitivity and Endpoint Underdetermination in Neural Network Training (DOI: 10.5281/zenodo.19118807) Paper 11: Post-Hoc Disclosure Is Not Runtime Proof: Model Identity at Frontier Scale (DOI: 10.5281/zenodo.19216634) Paper 12: Family-Dependent Response to Reasoning Distillation Across Structural and Functional Identity Layers (DOI: 10.5281/zenodo.19298857) Paper 13: Safety-Alignment Removal as a Model-Identity Failure — Structural Evidence from Published Weight-Level Mutation Checkpoints (DOI: 10.5281/zenodo.19383019) Technical Note: Agent Identity Is Not Model Identity (DOI: 10.5281/zenodo.19240883) Technical Note: Gap Invariance: Why PPP Measurements Are Domain-Independent by Construction (DOI: 10.5281/zenodo.19275524) Technical Note: Measured Model Substitution Under Valid Agent Credentials (DOI: 10.5281/zenodo.19342848) Technical Note: Artifact Identity Is Not Runtime Identity — Trustfall Lite and the Boundary of File-Level Model Verification (DOI: 10.5281/zenodo.20019127) Formal Verification Stack for Neural Network Structural Identity (IT-PUF Coq Proofs) (DOI: 10.5281/zenodo.18930621) Copyright (c) 2026 Anthony Ray Coslett / Fall Risk AI, LLC. All Rights Reserved. Confidential and Proprietary. Patent Pending (Applications 63/982,893, 63/990,487, 63/996,680, 64/003,244).
Tanusree Sharma, Yujin Potter, Jongwon Park, Yiren Liu · 9 authors
A major criticism of AI development is the lack of transparency, such as, inadequate documentation and traceability in its design and decision-making processes, leading to adverse outcomes including discrimination, lack of inclusivity and representation, and breaches of legal regulations. Underserved populations, in particular, are disproportionately affected by these design decisions. Furthermore, traditional social science techniques such as interviews, focus groups, and surveys struggle to adequately capture user needs and expectations in the digital era, due to their inherent limitations in deliberation, consensus-building, and providing consistent insights. We developed a democratic decision framework utilizing Decentralized Autonomous Organization (DAO) to enable underserved groups to deliberate and reach a consensus on key AI issues. To assess our proposed democratic decision mechanism, we conducted a case study on updating AI model specification based on diverse stakeholders input. We focus on reducing stereotypical biases in text-to-image systems, particularly gender bias in image generation from text prompts. We designed and experimented various governance configurations, including decision aggregation schemes and decision power, to examine how democratic processes could guide updates to AI model. Through a 2 × 2 experimental design, we tested various aggregation schemes (ranked vs. quadratic) and decision power distribution (equal vs. 20/80 differential) in a randomized online experiment (n=177) with participants from the global south and people with disabilities, to study how the varying governance mechanisms impact people's perceptions of the decision-making processes and resulting output of the AI Model specification. Our results indicate that despite their diverse backgrounds, participants showed convergence in deliberations on several aspects, including user control over image generation, multiple output options for user selection, and the social appropriateness and accuracy of generated images. Our study underscores the importance of use of appropriate governance in democratic decision-making in AI alignment. Notably, the combination of quadratic preference aggregation method which gives minorities more voice and equal decision power distribution, was perceived as a fairer and democratic approach.
Open access
Ethics and Social Impacts of AI
Artificial Intelligence in Healthcare and Education
Modern insurance organizations have adopted artificial intelligence in narrow, task-specific roles, resulting in fragmented systems that optimize isolated functions without fundamentally reshaping the underwriting and claims lifecycle. This “incrementalism” yields a human-default, sequential process plagued by structural bottlenecks, inconsistent risk evaluation, and limited transparency. This paper introduces NEXUS (Next-Generation Executive Underwriting and Settlement Intelligence), a framework to re-architect insurance as an AI-native system. NEXUS transitions AI from a peripheral tool to the primary orchestrator of end-to-end processes, conceptualizing the insurance lifecycle as a conversational, agent-orchestrated workflow. It is realized through a unified conversational interface that coordinates a decentralized ecosystem of specialized, collaborative AI agents each responsible for domain-specific reasoning such as geospatial risk assessment, financial verification, or medical outcome analysis. The central innovation is the Truth Score Engine (TSE), a governance-first aggregation mechanism that non-linearly synthesizes agent outputs by weighting evidentiary provenance, confidence estimates, and cross-agent consistency. The TSE governs decisions via a Three-Tiered Confidence Protocol: • High Confidence (&gt;90%) validates outcomes for immediate human sign-off without re-verification; • Medium Confidence (60-90%) routes decision summaries for targeted human review of specific flags; • Low Confidence (&lt;60%) escalates cases as ‘’Risky,’’ reverting to traditional manual investigation. This protocol yields a single, auditable decision artifact while preserving full traceability of the reasoning pathway. By embedding multi-agent coordination, contextual awareness, and tiered governance at the architectural level, NEXUS demonstrates a scalable pathway toward adaptive, transparent insurance systems. It ensures precision, combats fraud, and dramatically reduces settlement time, positioning AI-native governance as a foundational requirement for deploying trusted, autonomous decision-making in high-stakes financial domains.
We present Viturka, a blockchain architecture that replaces wasteful proof-of-work mining with productive federated learning. The core innovation is Proof of Credibility (PoC): a consensus mechanism where block production probability is determined by accumulated reputation from validated AI contributions rather than computational hash power or financial stake. Viturka leverages recent breakthroughs in Zero-Knowledge Machine Learning (ZKML) to achieve cryptographic verification of model training. Validators generate zero-knowledge proofs attesting to correct training execution, enabling instant on-chain verification without trusted intermediaries or statistical consensus mechanisms. By integrating frameworks like EZKL and Lagrange's DeepProve with GPU-accelerated proving via the Icicle library, validation that previously required hours of recomputation now produces mathematical proofs verifiable in milliseconds. Participants earn credibility by contributing quality training data or validating others' contributions. Only the top 10 highest-credibility validators can participate in validation rounds, with mandatory cooldown periods ensuring rotation. The system uses a temporal commit-reveal scheme for data contributions combined with ZK proofs for validation—fake contributions result in permanent bans, while fraudulent validation is mathematically impossible. This creates infrastructure for training AI models on distributed data without central coordination, with economic incentives aligned toward data quality rather than raw computation. Applications range from commercially valuable use cases like DeFi credit scoring—which could unlock over $100B in overcollateralized capital—to public-good AI for rare diseases, minority languages, and environmental monitoring.
Protecting the intellectual property of large language models (LLMs) is a critical challenge due to the proliferation of unauthorized derivative models. We introduce a novel fingerprinting framework that leverages the behavioral patterns induced by safety alignment, applying the concept of refusal vectors for LLM provenance tracking. These vectors, extracted from directional patterns in a model's internal representations when processing harmful versus harmless prompts, serve as robust behavioral fingerprints. Our contribution lies in developing a fingerprinting system around this concept and conducting extensive validation of its effectiveness for IP protection. We demonstrate that these behavioral fingerprints are highly robust against common modifications, including finetunes, merges, and quantization. Our experiments show that the fingerprint is unique to each model family, with low cosine similarity between independently trained models. In a large-scale identification task across 76 offspring models, our method achieves 100\% accuracy in identifying the correct base model family. Furthermore, we analyze the fingerprint's behavior under alignment-breaking attacks, finding that while performance degrades significantly, detectable traces remain. Finally, we propose a theoretical framework to transform this private fingerprint into a publicly verifiable, privacy-preserving artifact using locality-sensitive hashing and zero-knowledge proofs.
Walter Kurz, Michel Malara, Wojtek Stricker, Eva Albrecht
The objective of this study is to define a compliance-first, conceptually generalisable architecture for a multi-agent artificial intelligence platform integrated with distributed ledger technology, designed to be domain-, deployment-, and vendor-agnostic. It addresses a persistent shortcoming in current AI deployments, where compliance is often treated as a secondary concern, applied retroactively through prompt engineering rather than embedded within the foundational design. The proposed model encodes regulatory, governance, and ESG requirements into an objective-under-constraints framework, ensuring that all specialised agents operate within legally admissible and verifiably auditable parameters prior to any domain-specific implementation. A DAG-based verification layer is incorporated to enable scalable, low-latency, and cost-efficient operation while preserving evidentiary integrity. The analysis evaluates the feasibility of this conceptual model to support sustainable, rapid-deployment vertical applications without inducing vendor lock-in, preserving operational neutrality, and ensuring environmental accountability. The findings suggest that integrating compliance, ESG metrics, and agent specialisation at the architectural level provides a transferable foundation for cross-domain AI–DLT infrastructures.
Financial Generative Pre-trained Transformers (FinGPT) with multimodal capabilities are now being increasingly adopted in various financial applications. However, due to the intellectual property of model weights and the copyright of training corpus and benchmarking questions, verifying the legitimacy of GPT's model weights and the credibility of model outputs is a pressing challenge. In this paper, we introduce a novel zkFinGPT scheme that applies zero-knowledge proofs (ZKPs) to high-value financial use cases, enabling verification while protecting data privacy. We describe how zkFinGPT will be applied to three financial use cases. Our experiments on two existing packages reveal that zkFinGPT introduces substantial computational overhead that hinders its real-world adoption. E.g., for LLama3-8B model, it generates a commitment file of $7.97$MB using $531$ seconds, and takes $620$ seconds to prove and $2.36$ seconds to verify.