AI-Resistant Physical Security through Deterministic Plasmonic Encoding

Recent advances in artificial intelligence have significantly improved the ability to analyze and replicate physical security tags, creating an urgent need for security platforms that remain robust against data-driven attacks. Here, we present a magnetoplasmonic platform that combines deterministic plasmonic encoding with fundamentally non-invertible decoding. Magnetically responsive plasmonic nanoparticles are aligned by programmed three-dimensional magnetic fields and permanently fixed through photopolymerization, converting magnetic-field information into polarization-dependent plasmonic color patterns with high reproducibility. The information capacity is further expanded by spatially multiplexing multiple magnetic-field configurations within a single pixelated plasmonic tag. This magnetoplasmonic keychain integrates magnetic configurations with indexed spatial assignments, generating a physical key space exceeding O(44) while preserving straightforward optical readout when the complete key is available. Unlike conventional optical security tags, the proposed platform is intentionally designed so that the inverse mapping from optical patterns to magnetic configurations becomes ill-posed once the spatial key is removed. Under these conditions, different magnetic configurations produce indistinguishable optical observations, preventing unique reconstruction of the encoded information. Consequently, supervised deep-learning models fail to decode the security tags beyond random guessing, whereas accurate reconstruction is immediately achieved when the correct keychain is provided. This work introduces a physically enforced security framework in which information protection originates from the non-invertibility of the underlying physical process rather than computational complexity. The proposed strategy establishes a scalable route toward AI-resistant physical cryptography for next-generation authentication, anti-counterfeiting, and secure information storage.