The Internet of Things (IoT) represents a vast global network interconnecting a wide array of smart devices. The rapid growth of the IoT has significantly expanded the number of connected endpoints, leading to new and complex challenges in network security, such as eavesdropping, weak authentication, and malicious payloads. Many IoT networks rely on lightweight protocols like MQTT, which, although efficient, are susceptible to a diverse range of cyber threats. Traditional intrusion detection methods struggle to keep pace with the evolving attack patterns due to the constrained resources of IoT devices and the dynamic nature of network traffic. This study introduces a Transformer-based IoT Traffic Intrusion Detection (TITID) model, which leverages the Transformer architecture to improve classification accuracy. By integrating feature embeddings and an Intra-Sequence Attention mechanism, the proposed model significantly enhances the detection of cyber threats within IoT networks. Experimental evaluations on the MQTT-IoT-IDS2020 dataset demonstrate the superiority of TITID over existing approaches, underscoring its potential for enhancing IoT security.