Holo-Code: A Robust and Statistically Lossless Holographic Watermarking Framework for AIGC in Cloud-Edge IoT

The proliferation of AI-Generated Content (AIGC) within cloud-edge IoT ecosystems necessitates robust provenance mechanisms to ensure copyright protection and forensic traceability. However, the open distribution mechanism of AIGC exposes proprietary assets to risks of unauthorized replication and misuse. While watermarking offers a solution, existing schemes face a critical dilemma in IoT environments: they are either computationally expensive for edge deployment or structurally fragile against the dual threats of unpredictable IoT transmission failures and post-distribution image manipulations. To overcome these limitations, we propose Holo-Code, a robust and statistically lossless watermarking architecture based on a decoupled channel-coding paradigm. We introduce a holographic topological transformation to globally disperse watermark information across the latent representation. This mechanism mathematically converts structural data loss into uniform random erasures, which are robustly recovered via concatenated error correction coding. Furthermore, Holo-Code incorporates a distribution whitening mechanism that ensures the embedded signal is statistically indistinguishable from the standard Gaussian prior, thereby preserving the high fidelity of generated images. Extensive experiments validate that Holo-Code achieves a superior trade-off between robustness, quality, and efficiency. The framework guarantees reliable traceability across multiple generative models even under extreme composite degradations, seamlessly surviving heavy quantization from edge fallback mechanisms combined with severe structural occlusion. Moreover, as a training-free intervention, Holo-Code executes the complete forensic extraction on standard CPUs and achieves highly efficient identity retrieval against a database of one million users, establishing it as a practical solution for real-time AIGC provenance in large-scale IoT networks.