AI / ML / Autonomous Systems — 2026-03-18
Pentagon Plans Secure Environments for AI Training on Classified Military Data
A senior defense official confirmed the Pentagon is building secure environments for AI firms to train military-specific model variants on classified data, including surveillance reports and battlefield assessments. This goes beyond current use to embedding classified intelligence directly into model weights. The DoD signed agreements with OpenAI and xAI; Anthropic was expelled from the military supply chain in early March.
Analysis
Prior digests tracked Anthropic's DODsupply chain risk designation (March 6) and Pentagon AI use in Iran target analysis (March 11). The classified training disclosure represents a significant escalation -- from using commercial models to query classified data, to permanently embedding classified intelligence into model weights.
Prior digests tracked Anthropic's DOD