<p>In response to the problems of low accuracy, low efficiency, and insufficient data security in building information modeling models, an automation scheme based on deep learning and differential privacy is introduced to improve the efficiency and security of construction projects. Firstly, this article utilized residual network and long short-term memory models to extract features from the spatial structure and temporal information of the building information modeling model. Secondly, it used the Adam optimizer for gradient descent to optimize network weights; then, for the evaluation of geometric accuracy and data integrity, this article adjusted the number of network layers and convolution kernel size. At the same time, this article integrated a differential privacy framework into the model, dynamically adjusted the privacy budget, and finally evaluated the building information modeling model for automated generation and protection. The research results indicated that the building information modeling model had an accuracy of 0.94 in aligning the edges of the structural framework, and the internal partition wall angle deviation was only 0.10. Its error in the area covered by internal partitions and floors was only 0.04, with a success rate of 0.82 for data leakage risk, a detection rate of 0.91, and a response time of 0.09&#xa0;s. The model can ensure the security of critical information while achieving high accuracy and efficiency.</p>