The Bank of Italy has deployed artificial intelligence to detect microscopic printing errors in euro banknotes, according to a technical study released on June 4 that positions the institution at the forefront of automated currency quality control within the Eurozone. The system—a neural network architecture known as siamese networks—now supports human inspectors in spotting flaws that conventional machinery might miss, delivering precision rates above 95% in live production trials.
Why This Matters
• Quality assurance redefined: The Bank of Italy processed a test sample of 446 banknote images representing 21 distinct defect categories, achieving a 1.0 accuracy score on obverse checks and 0.95 on reverse surfaces.
• Jobs remain human-led: While automation streamlines initial screening, final approval authority stays with trained operators—a crucial distinction as Italy strengthens its currency quality standards while preserving specialized employment in the sector.
• Eurozone benchmark: The model aligns with the European Central Bank's common quality framework, potentially setting a template for printing facilities across member states.
How the Technology Works
The neural approach differs sharply from traditional inspection methods. Rather than programming a machine to recognize every conceivable flaw—a near-impossible task given variations in defect shape, severity, and location—the Bank of Italy's research team employed few-shot learning. That technique trains the network by comparing pairs of images: one pristine reference note and one suspect specimen.
Each half of the siamese architecture analyzes one image, converting it into a unique digital profile. The system then compares the two profiles and flags any significant difference. Because the twin components share the same parameters, the model adapts well even when encountering defect types absent from the training set—a critical advantage on a production line where anomalies evolve continuously.
The Bank of Italy published its findings in Working Paper No. 82 of the "Markets, Infrastructures, and Payment Systems" series, building on preliminary siamese-network experiments conducted in May 2023. Engineers credited the architecture's explainability feature—a visual overlay highlighting suspect regions—for winning operator trust. Inspectors can now focus on problem zones rather than scrutinizing every square millimeter manually, cutting review time without sacrificing rigor.
What This Means for Currency Integrity and Italian Quality Standards
Automated vision systems have screened banknotes in Italy for decades, yet the final quality gate remained a bottleneck staffed by specialists who assess type, extent, severity, and position of printing irregularities. Those parameters determine whether an entire production batch meets Eurosystem standards or gets downgraded to destruction.
By filtering out obvious passes and presenting only borderline cases to human experts, the AI layer accelerates throughput and reduces fatigue-related errors. The Bank of Italy emphasized in its paper that decision-making authority rests with people, framing the technology as augmentation rather than replacement. For Italians handling euros daily, this development means the banknotes in circulation will meet even stricter quality standards while maintaining the expertise of human specialists who understand currency authentication and fraud prevention.
The Bank of Italy has publicly committed to retaining its in-house banknote production function and negotiating workforce continuity clauses should organizational changes ripple through the Eurozone. That pledge reflects awareness that expertise in currency quality control carries strategic value both for Italy's economy and the broader Eurozone's confidence in the euro.
The Broader European Context
Italy's initiative unfolds against a backdrop of continent-wide interest in AI-driven cash management. The European Central Bank listed AI applications in banking—including fraud detection and credit assessment—as a supervisory priority for 2026–2028 and hosts regular workshops to map use cases. A Decision (ECB/2025/36) issued in October 2025 streamlined procedures for checking euro-note authenticity and fitness, hinting at procedural updates designed to accommodate machine-learning tools.
France's Banque de France convened an AI in Finance and Central Banking workshop in October 2025, exploring how monetary authorities can harness emerging tools for note integrity and quality assurance. Other central banks across the Eurozone are similarly exploring AI applications for currency management, positioning Italy's research as part of a coordinated European effort to strengthen the euro's quality and security.
Impact on Production Economics
Automated inspection delivers measurable cost and speed gains. Vision systems operate around the clock without breaks, scan every unit in real time, and generate structured data logs that feed continuous-improvement cycles. For a central bank managing multi-million-unit print runs, even fractional reductions in error rates or rework translate into significant savings.
Quality improvements also carry reputational weight. Public confidence in cash hinges on uniform appearance and robust security features; a batch of visibly flawed notes erodes trust and invites counterfeiting speculation. By tightening tolerances at the production stage, the Bank of Italy reinforces the euro's credibility—an intangible yet economically vital asset for Italy's place within the Eurozone.
The siamese-network model's ability to adapt to novel defects without exhaustive retraining further lowers lifecycle costs. Traditional machine-vision setups require engineers to script detection rules for each new flaw type, a labor-intensive process that delays deployment. Few-shot learning collapses that cycle, enabling inspectors to upload a handful of annotated examples and update the system within hours.
Looking Ahead
The Bank of Italy's research team acknowledged that precision on the reverse side—0.95—leaves room for refinement, particularly for edge cases involving subtle color shifts or micro-perforations. The study suggests pathways for future enhancement without speculating on specific technologies.
Regulatory frameworks will also evolve. As AI assumes a larger role in currency lifecycle management—from design validation through destruction—supervisory bodies must establish audit trails, bias checks, and fail-safe protocols. The ECB's 2026–2028 agenda suggests that oversight will extend beyond commercial banks to central-bank operations, formalizing standards for algorithmic transparency and human-in-the-loop safeguards.
For Italy's workforce in banking and currency production, the trajectory hinges on reskilling initiatives that transition inspectors from manual scrutiny to AI supervision—monitoring model performance, annotating edge cases, and escalating anomalies. The Bank of Italy appears committed to ensuring that productivity gains do not come at the expense of employment security, a balance critical for maintaining public confidence in Italy's role as a steward of the euro.
