온디바이스 AI 환경의 하드웨어 기반 콘텐츠 출처 인증 프레임워크 연구 = A Study on Hardware-Based Content Provenance Authentication Framework for On-Device AI

다국어 초록 (Multilingual Abstract)

The proliferation of generative AI and on-device AI has enabled edge devices such as smartphones, smart glasses, and IoT devices to generate, capture, and edit content locally. However, software-based security mechanisms face structural limitations including OS-level bypass risks, cryptographic overhead under constrained resources, and offline authentication gaps. This study proposes HW-CPA (Hardware-based Content Provenance Authentication), a hardware-oriented framework comprising four layers: NPU-based content classification, TEE-based C2PA manifest generation, Hardware Root of Trust-based device identity assurance, and offline hash chain-based deferred verification. Threat modeling, commercial technology mapping, and a software emulation prototype confirm consistent tamper detection and acceptable signing/verification latency for consumer-grade environments. This work provides a foundation for future SoC-level integration toward trustworthy generative AI content.