Pentagon Plans Secure Enclaves for AI Companies to Train Models on Classified Intelligence Data

The Pentagon will allow commercial AI firms to train government-specific models directly on classified data inside accredited secure enclaves—a first for the IC. OpenAI and xAI have agreements to operate in classified settings; Anthropic is excluded following its supply chain risk designation. The initiative, accelerated by a January Hegseth memo, pairs LLMs with intelligence reports and operational planning data. The selective vendor access reshapes the competitive landscape for IC AI contracts.