1. The Counterfeiting Problem at Scale
QR code counterfeiting has evolved significantly beyond simple URL substitution. Sophisticated counterfeiters now produce physically accurate replicas of brand-specific QR codes — including custom colors, embedded logos, and serialized payloads — using commercial inkjet printers and stolen product databases. In 2025, the OECD estimated that QR-related product counterfeiting contributed to approximately 3.3% of global trade fraud by volume. The challenge for authentication systems is distinguishing a legitimate QR label from a counterfeit one that encodes a valid-looking payload.AI-based authentication approaches operate on two distinct layers: payload verification (checking that the encoded data matches a known-good value in a secure registry) and substrate forensics (analyzing the physical printing characteristics of the QR label itself to detect reproduction artifacts). The combination of both layers is significantly more robust than either alone.
2. Payload Authentication with Server-Side Verification
The most widely deployed anti-counterfeit QR architecture uses cryptographically signed payloads. Each authentic QR code encodes a URL containing a unique serial number and a digital signature generated by the brand's private key using ECDSA (Elliptic Curve Digital Signature Algorithm) over the NIST P-256 curve. When the QR is scanned, the landing page server verifies the signature against the brand's public key. A valid signature confirms that the URL was generated by the brand's signing infrastructure, not forged.This approach is standardized in GS1's Digital Link framework (GS1 DL 1.2), which specifies how product identifiers (GTIN, serial number, lot number) and cryptographic verification data are encoded into QR payloads. Major pharmaceutical manufacturers use GS1 DL with ECDSA-signed serial numbers to comply with EU Falsified Medicines Directive (FMD) serialization requirements.
3. AI Substrate Forensics: Print Artifact Analysis
Legitimate QR labels are produced by industrial printing processes (offset lithography, flexographic printing, thermal transfer) that have characteristic microscopic ink deposition patterns. Counterfeit labels printed on consumer inkjet or laser printers exhibit different halftone dot structures, ink bleeding patterns, and edge sharpness profiles. A convolutional neural network trained on macro-photography images of authentic versus counterfeit labels can classify the print substrate with greater than 96% accuracy, providing a forensic authentication layer that operates independently of the payload content.This technology is deployed in smartphone-based authentication apps using the phone's rear camera at 2× to 5× optical zoom. The CNN model (typically MobileNetV3 or EfficientNet-Lite for on-device inference) analyzes the microscopic texture of the QR modules and surrounding substrate, then returns an authenticity confidence score within 500 milliseconds. Kezzler, Systech, and Authentix offer commercial implementations of this approach.
4. Anomaly Detection for Scan Pattern Analysis
Beyond individual code inspection, AI anomaly detection models analyzing aggregate scan patterns can identify counterfeiting operations at a network level. A legitimate product typically generates a geographic scan pattern consistent with its distribution chain: factory scan in China, port scan in Rotterdam, retailer scan in Paris, consumer scan in a French household. An anomalous pattern — such as the same serial number being scanned simultaneously in three different countries, or a serial number appearing in a retail scan before it appears in a factory scan — signals duplication.Isolation Forest, Autoencoder, and LSTM-based anomaly detectors trained on historical supply chain scan sequences can flag these violations in near real-time, triggering alerts in the brand's supply chain visibility platform (SAP Integrated Business Planning, Oracle SCM Cloud) and automatically deactivating the compromised serial number to prevent further fraudulent use.