Steganography is the art of hiding secret data within cover media for covert communication. Recent studies have shown that Fixed Neural Network Steganography (FNNS) exhibits significant practicality in real-world applications due to its ability to achieve high steganographic performance without the need for network training. However, the stego images generated by existing FNNS methods often exhibit high distortion, making them easily detectable by steganalysis tools, thereby limiting their security. To address this issue, we propose a high-security multi-channel hybrid perturbation approach for Fixed Neural Network Steganography, referred to as MHPA. To enhance the expressive capability of the perturbations, we introduce a Hybrid Nonlinear Perturbation strategy to adapt to various complex steganographic conditions. Meanwhile, we propose a multi-channel sampling strategy to reduce the visual impact of the perturbations on the cover images during the embedding process, while also preserving more secret information. Additionally, when iteratively updating the perturbations, we utilize the gradients from multiple steganalyzers to guide the updates. Extensive experimental results indicate that our method not only generates high-quality stego images and effectively recovers images but also significantly enhances security. Specifically, compared to classical FNNS methods, the stego images produced by our approach show an improvement in image quality of 21.51 dB. When resisting mainstream steganalyzers such as SRNet and YeNet, our method outperforms state-of-the-art (SOTA) FNNS methods by 47.6% and 15.90%, respectively. The code is available at https://github.com/AlexJakin/MHPA_FNNS .