War Department Signs Agreements to Deploy AI on Classified Networks

According to the War Department press release, the department has entered into agreements with eight frontier AI companies, SpaceX, OpenAI, Google, NVIDIA, Reflection, Microsoft, Amazon Web Services, and Oracle, to deploy advanced AI capabilities on its classified networks for lawful operational use. The announcement says these capabilities will operate on Impact Level 6 (IL6) and Impact Level 7 (IL7) network environments, and that the effort supports the Department's AI Acceleration Strategy. The release states that the department's official AI platform, GenAI.mil, has been used by more than 1.3 million Department personnel, generating "tens of millions" of prompts and deploying "hundreds of thousands" of agents in five months, per the press release.

Breaking Defense reports that some companies were already on contract while others are finalising terms, and that Oracle was added to an earlier published list that initially named seven firms. The Oracle press release confirms a separate agreement between Oracle and the War Department and includes direct statements from Oracle executives. Multiple outlets, including CNBC and Army Times, report that Anthropic is not on the approved list and describe Anthropic as remaining subject to restrictions or being excluded from government work.

The War Department text frames the deployment as integrating "secure frontier AI capabilities" into IL6 and IL7 to streamline data synthesis, elevate situational understanding, and augment warfighter decision-making. The release also states the Department is building an architecture intended to avoid vendor lock and to provide a "diverse suite" of AI capabilities from across the domestic technology stack. Oracle's announcement highlights openness, interoperability, and embedding generative and agentic AI into enterprise software as priorities in its agreement; Oracle's PR includes quotes from Kim Lynch and Emil Michael on operational advantage and secure infrastructure.

Editorial analysis - technical context: Deploying commercial frontier models inside classified enclaves creates several engineering and operational demands that are familiar to large-scale government and regulated deployments. These include strict data isolation, vetted supply-chain components, vetted model provenance, hardened inference stacks, and audited telemetry to meet IL6 and IL7 controls. Systems integrators and cloud providers typically implement containerization, hardware attestation, and offline model vetting to bridge commercial models with classified security baselines.

Industry context: Public reporting frames this as a continuation of Defense Department efforts to operationalise commercial AI for warfighting, intelligence, and enterprise operations. For the commercial AI ecosystem, having multiple major vendors cleared for classified environments expands procurement pathways for defence customers while simultaneously concentrating scrutiny on supply-chain risk and compliance. The reported exclusion of Anthropic, covered by CNBC and other outlets, underscores how regulatory and supply-chain assessments are shaping which providers can access sensitive government networks.

For practitioners: Access to classified enclaves does not obviate engineering work. Teams integrating these capabilities should plan for integration testing against IL6/IL7 security baselines, model evaluation under red-team scenarios, evidence collection for auditability, and performance benchmarking in constrained or air-gapped contexts. Contractual differences between providers, and the fact that some deals were reported as already under contract while others are still being finalised, imply variance in delivery timelines and integration support across vendors, per Breaking Defense reporting.

For practitioners and observers, track four indicators: 1) vendor onboarding documents and hardening guidance released by the War Department or participating vendors; 2) technical integration patterns published for GenAI.mil and any interoperability specifications; 3) any formal definitions or guidance that clarify IL7 as used operationally; and 4) procurement or supply-chain assessments that detail why certain vendors, such as Anthropic, remain excluded according to reporting from CNBC and Army Times.

Editorial analysis: Overall, the announcements mark a notable step toward routine use of commercial frontier models in classified settings, but they also mirror longstanding engineering and governance challenges that accompany moving commercial AI into highly constrained, regulated environments. Practitioners should expect significant systems engineering and compliance work even after vendor approval appears on paper.