Pentagon Approves Eight AI Vendors For IL7 Classified Networks

The U.S. military is expanding its generative AI capabilities across highly restricted network environments. The Pentagon signed agreements with eight technology providers to deploy foundation models and AI tools directly onto classified infrastructure. The newly authorized vendor list shifts the defense sector’s approach to operational guardrails and model constraints.

Classified Infrastructure Integration

The new deployment agreements grant vendor systems access to Impact Level 6 (IL6) networks, which handle “Secret” classified data, as well as the semi-official Impact Level 7 (IL7) designation reserved for the most sensitive intelligence systems. These integrations pipeline directly into GenAI.mil, the central platform managing the military’s machine learning workloads.

Adoption across defense personnel has scaled rapidly. In five months, more than 1.3 million personnel have accessed the GenAI.mil environment. The system currently manages tens of millions of prompts and has deployed hundreds of thousands of autonomous agents for data synthesis and situational understanding across complex multi-agent systems. For engineers evaluating and testing AI agents at scale, the military’s infrastructure represents one of the largest active deployments of agentic architecture.

Authorized Vendors and Operational Standards

The finalized roster includes a mix of proprietary model developers, cloud infrastructure providers, and specialized startups.

Vendor	Known Platform/Model Scope
Amazon Web Services	Infrastructure and AI services
Google	Gemini models
Microsoft	Infrastructure and model deployment
Nvidia	Hardware and software integration
OpenAI	Proprietary foundation models
Oracle	Infrastructure services
Reflection	Nvidia-backed startup models
SpaceX	xAI Grok

These deployments operate under a new contractual standard defined as “any lawful operational use.” This requirement explicitly overrides the default safety guardrails embedded in commercial APIs, ensuring that military operators face no programmatic restrictions when using the models for complex operational planning or intelligence analysis.

The Anthropic Exclusion

The multi-vendor agreement directly replaces the military’s reliance on Anthropic. The pivot follows a highly publicized dispute in February 2026, when Anthropic refused to disable safety mechanisms in its Claude models that block usage for mass domestic surveillance and fully autonomous lethal weapons.

The refusal prompted President Trump’s administration to mandate the removal of Anthropic systems from federal agencies. Secretary of Defense Pete Hegseth subsequently classified Anthropic as a “national security risk” and “supply chain risk.” Anthropic is currently disputing the designations in court.

The operational shift has triggered internal friction at participating companies. Approximately 600 Google employees have formally protested the deployment of Gemini models on classified military infrastructure. Despite this, Pentagon CTO Emil Michael indicated that establishing a broad, diversified vendor base is necessary to avoid over-reliance on any single corporate partner.

If you build AI applications for federal defense contractors, the GenAI.mil requirements establish a new baseline for acceptable use policies. Models integrated into IL6 and IL7 environments must support unfiltered operational commands, meaning standard commercial API guardrails will need to be stripped or systematically bypassed to meet federal deployment standards.