The paper introduces GS-Hider, a novel steganography framework designed for 3D Gaussian Splatting (3DGS), a state-of-the-art method for 3D scene reconstruction and novel view synthesis. Given the significant computational cost and the need to protect the copyright and privacy of 3D assets, the authors propose GS-Hider to embed and extract messages from 3DGS point cloud files without compromising the original scene's fidelity or security. Key contributions of GS-Hider include: 1. **Security and Fidelity**: GS-Hider ensures robust security and high fidelity by using a coupled secured feature attribute and parallel decoders. 2. **Large Capacity and Versatility**: It can hide multiple 3D scenes or images into a single 3D scene, enhancing adaptability and engagement. 3. **Efficiency**: The method maintains real-time rendering capabilities, achieving a rendering speed of 45 fps. The proposed framework replaces the spherical harmonics coefficients in 3DGS with a coupled secured feature attribute, which is then rendered using a coupled feature rendering pipeline. Two parallel decoders, a scene decoder and a message decoder, are used to disentangle the original scene and the hidden message. Extensive experiments demonstrate the effectiveness of GS-Hider in terms of security, robustness, fidelity, and flexibility. The method is evaluated on the Mip-NeRF360 dataset, showing superior performance in hiding messages while maintaining high-quality rendering.The paper introduces GS-Hider, a novel steganography framework designed for 3D Gaussian Splatting (3DGS), a state-of-the-art method for 3D scene reconstruction and novel view synthesis. Given the significant computational cost and the need to protect the copyright and privacy of 3D assets, the authors propose GS-Hider to embed and extract messages from 3DGS point cloud files without compromising the original scene's fidelity or security. Key contributions of GS-Hider include: 1. **Security and Fidelity**: GS-Hider ensures robust security and high fidelity by using a coupled secured feature attribute and parallel decoders. 2. **Large Capacity and Versatility**: It can hide multiple 3D scenes or images into a single 3D scene, enhancing adaptability and engagement. 3. **Efficiency**: The method maintains real-time rendering capabilities, achieving a rendering speed of 45 fps. The proposed framework replaces the spherical harmonics coefficients in 3DGS with a coupled secured feature attribute, which is then rendered using a coupled feature rendering pipeline. Two parallel decoders, a scene decoder and a message decoder, are used to disentangle the original scene and the hidden message. Extensive experiments demonstrate the effectiveness of GS-Hider in terms of security, robustness, fidelity, and flexibility. The method is evaluated on the Mip-NeRF360 dataset, showing superior performance in hiding messages while maintaining high-quality rendering.