Search for a command to run...
Artificial intelligence (AI) data centers host large-scale inference and training workloads that generate high volumes of legitimate network and compute traffic[1],[2]. This operational intensity makes it challenging to distinguish between normal usage spikes and malicious Denial-of-Service (DoS) or Distributed DoS (DDoS) attacks[3],[4]. Traditional defense mechanisms such as static firewalls, signature-based detection, and fixed rate limiting are insufficient against adaptive or AI-generated attacks[5],[6].This article proposes an autonomous defense framework based on AI agents that continuously analyze real-time telemetry from ingress gateways, service meshes, and workload schedulers. Each AI agent employs anomaly detection, pattern classification, and reinforcement learning[7] to identify abusive behaviors while preserving high-throughput AI services. Upon detecting an attack, agents collaboratively enforce mitigation strategies such as dynamic throttling, micro-segmentation, and workload isolation, minimizing collateral impact on active compute tasks[8]. The proposed approach establishes a foundation for self-defending, autonomous[9] AI infrastructures that can withstand increasingly sophisticated DoS threats.This article is structured as follows. Introduction outlines the rapid growth of AI data centers and the escalating threat of sophisticated DoS and DDoS attacks, while emphasizing the limitations of conventional firewalls and intrusion detection systems. This motivates the need for autonomous, AI-driven security mechanisms capable of real-time decision-making. Related work section reviews both traditional defense mechanisms and recent machine learning-based detection strategies. Proposed framework details the multi-layered hybrid architecture of the AI agents, their data flow model, and potential deployment configurations in cloud and edge environments. The detection and mitigation pipeline is elaborated through the classification models employed and an adaptive response strategy optimized for varying threat intensities. Paper concludes by highlighting the advantages of the autonomous AI-driven approach, acknowledging current limitations, and proposing directions for future enhancement.