With the popularization of the internet, cybercrime continues to increase, and traditional blacklist methods have difficulty in coping with new threats. To address this challenge, the authors propose a web domain name security access recognition algorithm based on bidirectional recurrent neural networks, aiming to more effectively combat domain name generation technology. This algorithm extracts richer semantic features at each layer through bidirectional recurrent neural networks to more accurately describe domain name features, thus effectively handling SGD problems in abnormal network traffic detection. The results show that compared with the other three algorithms, the model trained by HCA-BAGD has better performance and higher accuracy, successfully solving the problem of network security detection. This study emphasizes the importance of cybersecurity and emphasizes continuous innovation and the adoption of new technological tools to ensure the safe operation of the internet ecosystem, bringing new perspectives and solutions to research and applications in the field of cybersecurity.