From a smartphone photograph to a blockchain-anchored, AML-scored digital passport. Five integrated technology layers designed for institutional use and built to meet EU, US, and UK regulatory requirements. Additional IP protection through 20 patent claims filed.
Layer 1 — Visual Identity
Every artwork has a unique physical surface — micro-variations in brushstrokes, canvas weave, pigment texture, and craquelure that are as distinctive as a human fingerprint. MIRAS captures this identity from two photographs (front + back) taken with a standard smartphone.
The front image is processed by a DINOv2 self-supervised vision transformer that produces a 384-dimensional visual fingerprint — a compact vector capturing micro-features invisible to the human eye. The back image documents provenance markers (gallery labels, stamps, inventory numbers). Together they produce a SHA-256 cryptographic digest — the Visual DNA — which is non-reversible, non-forgeable, and can be re-verified at any time by re-photographing the artwork.
A high-quality print or reproduction fails verification because it cannot replicate the physical micro-texture of the original surface.
Automated Research
MIRAS doesn't just store provenance — it actively researches it. The system queries 10+ external databases via API, calculates a Provenance Completeness Index (PCI), and flags gaps automatically.
Historical Risk Windows are pre-configured: Nazi looting (1933–1945), Eastern European confiscations (1949–1990), Russian revolution (1917–1922), UNESCO cultural property (post-1970), Cuban confiscations (1953–1959). Any ownership gap during these periods triggers an automatic flag.
The system is intellectually honest: it never claims 100% provenance where it doesn't exist. Instead, it quantifies exactly what is documented and what remains unknown.
Layer 5 — Risk Intelligence
The industry's first computational AML risk model for art transactions. Two independent scoring engines analyse 15 risk variables with 7 contextual modifiers, producing a composite score (0–100) mapped to four risk bands.
The Asset & Context Risk Score (ARS) uniquely integrates data from the authentication layer — Visual DNA verification status, NFC authentication, and blockchain integrity — creating a cross-layer anomaly detection capability that no partnership of separate tools can replicate.
For scores above 55 (Red), the system auto-generates a Suspicious Transaction Report in goAML format (UNODC standard), ready for submission to any Financial Intelligence Unit.
Layer 2 — Physical Security
An NXP NTAG 424 DNA chip with AES-128 encryption is physically bound to the artwork. Each tap generates a unique, non-replayable cryptographic message (SUN/CMAC protocol) with a monotonically increasing counter.
The tag stores the SHA-256 digest of the Visual DNA, creating a bidirectional cryptographic link: the tag proves it belongs to this artwork, and the artwork proves it belongs to this tag.
Tamper-evident adhesive and a TagTamper wire loop detect any removal attempt. If the tag is compromised, the Visual DNA remains the source of truth — the artwork is always identifiable without the tag.
Trap Mode: A stolen NFC tag on a fake artwork triggers instant Visual DNA mismatch — and continues logging GPS, timestamp, and device data for law enforcement.
Layer 3 — Immutable Record
MIRAS resolves the tension between data confidentiality (GDPR) and regulatory auditability using a dual-layer architecture.
Private layer (Hyperledger Fabric): stores confidential records — personal data, transaction details, full provenance chain — with Role-Based Access Control. Supports GDPR Art. 17 right to erasure.
Public layer (Ethereum L2 / Polygon): stores only irreversible hashes and ERC-721 tokens. Any third party can verify the integrity of a record without accessing confidential data.
An Independent Audit Node verifies every anchor — third parties don't need to trust MIRAS to verify proof.
Security & Compliance
SOC 2 Type I certification targeted for Year 1. Penetration tested before launch. Infrastructure hosted on AWS multi-region EU (Frankfurt primary, Ireland secondary) for GDPR data residency.
GDPR by architecture: personal data encrypted in Data Vault (AES-256 + RBAC). Blockchain stores only irreversible hashes. Right to erasure without compromising audit trail integrity.
EU AI Act (Art. 13, 14): every risk score includes a Risk Decomposition Report with top-5 contributing variables and natural language explanation. Human-in-the-loop mandatory for Orange and Red decisions.
Regulatory coverage: AMLR 2024/1624 (EU), 6AMLD (EU), Art Market Integrity Act S.2400 (US), HMRC MLR 2017 (UK), Basel AML Principles, RAM Toolkit.
Peer-Reviewed
Cross-Layer Anomaly Detection for AML Compliance in the Art Market: A Multi-Engine Scoring Framework
Tullio Angheben (MIRAS.ART — MARGIN International LLC), Margherita Angheben & Ginevra Angheben (CBS International Business School)
The paper introduces a computational framework purpose-built for anti-money laundering compliance in the art market. The system employs two independent scoring engines comprising 15 proprietary variables and 7 contextual modifiers. Using weighted aggregation with dynamic context-sensitive parameters, it produces composite risk scores that enable compliance with EU Regulation 2024/1624 with computational precision rather than subjective judgment.
Read the full paper on SSRN →AMLR 2027 is coming. The institutions that move first will set the standard. Technology is the enabler — being the institutional standard is the moat.
Request Early Access →