Modern networks generate large volumes of traffic that must be continuously monitored to identify malicious activity. Traditional signature-based intrusion detection systems struggle to identify novel or evolving attacks and often produce a high rate of false alerts. This paper presents a framework for AI-powered cyber threat detection based on the analysis of network traffic. Traffic flows are captured, pre-processed, and represented as feature vectors describing packet, flow, and behavioral characteristics. Machine learning and deep learning models are trained to distinguish normal traffic from malicious activity and to classify detected threats into categories such as denial-of-service, port scanning, brute-force access attempts, and malware communication. The framework emphasizes a repeatable pipeline covering data collection, feature extraction, model training, real-time inference, and alert generation. The proposed approach provides a basis for building adaptive, data-driven intrusion detection systems capable of identifying both known and previously unseen threats.