@techreport{akhavain-moussa-ai-network-01,
    number =    {draft-akhavain-moussa-ai-network-01},
    type =      {Internet-Draft},
    institution =   {Internet Engineering Task Force},
    publisher = {Internet Engineering Task Force},
    note =      {Work in Progress},
    url =       {https://datatracker.ietf.org/doc/draft-akhavain-moussa-ai-network/01/},
    author =    {Arashmid Akhavain and Hesham Moussa},
    title =     {{AI Network for Training, Inference, and Agentic Interactions}},
    pagetotal = 37,
    year =      2025,
    month =     nov,
    day =       2,
    abstract =  {Artificial Intelligence (AI) is rapidly transforming industries and everyday life, fueled by advances in model architectures, training paradigms, and data infrastructure for generation and consumption. Today, machine learning models are embedded in many of our daily activities, ranging from simple classification systems to advanced architectures such as large language models (LLMs) like ChatGPT, Claude, Grok, and DeepSeek. These models highlight the transformative potential of AI across diverse applications—from productivity tools to complex decision-making systems. However, the effectiveness and reliability of AI depend on two foundational processes: training and inference. Each process introduces unique challenges related to data management, computation, connectivity, privacy, trust, security, and governance. In this draft, we introduce the Data and Agent Aware-Inference and Training Network (DA-ITN)—a unified, intelligent, multi-plane network architecture designed to address the full spectrum of requirements needed to enable AI networks. DA-ITN provides a scalable and adaptive infrastructure that connects AI clients, data providers, model providers, agent providers, service facilitators, and computational resources to support end-to-end training, inference, and agentic interaction lifecycle operations. The architecture features dedicated control, data, and operations \& management (OAM) planes to orchestrate training, inference, and agentic services while ensuring reliability, transparency, and accountability. By outlining the key requirements of such an AI ecosystem and demonstrating how DA-ITN fulfills them, this document presents an architecture for the future of AI-native networking—an "AI internet"—optimized for AI learning, efficient inference, scalable deployment, and seamless agent-to-agent collaboration.},
}