Privacy Without Blindness: Ensuring AI Integrity, Relay Attack Resistance, and Crime Control in Offline Digital Euro Payments
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
IntroductionAI integrity is not a side issue in the Digital Euro — it is central to trust, fraud prevention, and lawful execution. Relay attacks, logic tampering, and fraudulent offline abuse should be blocked at the execution layer, not merely detected afterward. A fail-closed cryptographic architecture can help ensure that the Digital Euro remains privacy-preserving while still supporting targeted and lawful crime control. This is how the EU can build a Digital Euro that is secure against AI manipulation, resistant to relay attacks, and stronger on fraud prevention, while preserving proportionality and avoiding surveillance-based enforcement models. What This Paper Is About ? As the EU moves toward a digital euro and AI increasingly drives payment authorization decisions — fraud scoring, risk assessment, dynamic transaction limits, behavioral verification — a critical question remains unaddressed: how do we ensure that AI logic executing payment decisions offline is the same logic that was approved, audited, and deemed compliant? This article presents a cryptographic execution control architecture that enforces AI integrity at the moment of offline payment execution, without requiring network connectivity, identity tracking, or post-hoc fraud detection. It is designed to be directly applicable to the digital euro's offline payment requirements and to the EU AI Act's governance objectives for high-risk automated decision systems operating in financial contexts. The Problem Offline payments — whether via contactless cards, mobile wallets, secure elements, or a future digital euro — must authorize value transfer without a live connection to any central server. Today, this authorization relies on a combination of physical proximity (NFC signal presence), timing constraints, hardware counters, and transaction limits hardcoded into secure elements. Now add AI to this environment. AI-driven payment systems are already making real-time authorization decisions: should this transaction be approved? Is this behaviour consistent with the cardholder's profile? Does this transaction exceed a dynamic risk threshold? These AI models are increasingly embedded in payment devices, terminals, and secure elements themselves. The problem is this: once a device goes offline, there is no mechanism to verify that the AI logic making the payment decision is the logic that was originally approved. The model could have drifted. It could have been tampered with. It could have been substituted entirely. The secure element signs whatever it is told to sign. The terminal accepts whatever the device presents. Nobody checks whether the decision logic itself is authorised. This creates three cascading failures: First, relay attacks become possible because security depends on signal proximity, not execution authority. An attacker who relays NFC signals can trigger payment execution remotely — the system has no way to distinguish a legitimate local interaction from a relayed one, because proximity is treated as proof of legitimacy. Second, fraud prevention depends on identity tracking. To compensate for weak offline security, existing systems build behavioural profiles, track transaction histories, and correlate identities across sessions. This directly conflicts with EU data protection principles and the digital euro's stated privacy objectives. Third, AI integrity is unverifiable offline. If an AI model is making the authorization decision inside the device or terminal, and that model has been modified, no existing offline payment architecture can detect it before value transfer occurs. The damage is done before anyone knows. Why Existing Solutions Fail Existing offline payment architectures fail because they are built on a fundamentally flawed assumption: that physical presence equals legitimate execution authority. Proximity-based systems (NFC distance-bounding, timing checks, signal strength verification) assume that if a device is physically close to a terminal, the transaction is legitimate. Relay attacks shatter this assumption entirely. An attacker with two phones and a network connection can relay payment signals across continents in milliseconds. The terminal sees what appears to be a local device. The device sees what appears to be a local terminal. Neither can detect the relay. Secure element signing in current payment cards and devices provides cryptographic authentication of the device, but not verification of the decision logic. The secure element signs transaction requests — it does not evaluate whether the logic that produced the request is authorised. It is a cryptographic rubber stamp, not an execution authority gate. Post-execution fraud detection (risk scoring, pattern matching, anomaly detection) operates after value transfer has already occurred. In offline scenarios, this detection may be delayed by hours or days. The fraud has happened. The money has moved. Detection is not prevention. Identity-based fraud mitigation builds profiles, tracks behaviour, and correlates transactions across sessions to detect anomalies. This works, but at the cost of mass surveillance of payment behaviour — exactly what the EU digital euro regulation and GDPR seek to prevent. Offline CBDC prototypes (including current digital euro proposals) typically rely on hardware counters, value ceilings, and periodic reconciliation. They assume honest execution logic and accept residual fraud risk. None of them verify that the logic executing inside the device at the moment of payment is the logic that was approved. In summary: no existing offline payment system verifies execution logic integrity before value transfer. They verify identity, verify signals, verify proximity, verify signatures — but never verify the logic itself.
Fields of Study & Subjects
Blockchain Technology Applications and Security
Cryptography and Data Security
Advanced Authentication Protocols Security
Publication details
- Journal / source
- Zenodo (CERN European Organization for Nuclear Research)
- Publication date
- 30 March 2026
- Language
- EN