The Pentagon announced in June 2026 that it has signed agreements with eight major technology companies to deploy their artificial intelligence models on classified military networks through a new platform called GenAI.mil. The companies are OpenAI, Google, Microsoft, Amazon Web Services, Oracle, NVIDIA, SpaceX, and a firm called Reflection, according to the Department of Defense’s official announcement.
The rollout has already surfaced tensions over which companies made the cut, what classification level the platform actually serves, and how much human oversight will govern AI-assisted decisions in sensitive operations. The Pentagon describes the goal as enabling “lawful operational use” of commercial AI across military missions, spanning intelligence analysis, logistics planning, and other functions.
What the agreements cover
GenAI.mil is managed by the Pentagon’s Chief Digital and Artificial Intelligence Office (CDAO), the organization created in 2022 to centralize the military’s data and AI strategy. The platform is designed to give service members and analysts a common environment for testing, evaluating, and eventually using commercial AI models on government systems.
Google Cloud disclosed that CDAO selected its Gemini for Government product to help power GenAI.mil, according to a company press release. Google emphasized that its deployment meets Impact Level 5 (IL5) security standards, a tier that covers higher-sensitivity controlled unclassified information and certain national security workloads hosted in U.S. government cloud environments.
The other companies on the list bring different capabilities. OpenAI and Microsoft offer large language models and cloud infrastructure. Amazon Web Services and Oracle provide competing government cloud platforms. NVIDIA supplies the GPU hardware and software frameworks that underpin most modern AI training and inference. SpaceX and Reflection are less obvious picks: SpaceX’s inclusion likely relates to its Starshield satellite and secure communications work for the government, while Reflection is a smaller AI firm whose specific role has not been publicly detailed.
Seven companies or eight?
A notable discrepancy has emerged between the Pentagon’s own count and independent reporting. The Defense Department’s announcement names eight companies. The Associated Press, however, reported that the Pentagon reached deals with seven technology companies to run AI on classified systems.
No public explanation has resolved the gap. It is possible that one company signed an agreement but has not yet been cleared for operational deployment, or that one firm’s role differs enough from the others that the AP categorized it separately. The distinction is not trivial: each slot represents a multimillion-dollar defense relationship and access to some of the government’s most protected information systems.
The Anthropic injunction and vendor selection risks
Notably absent from the list is Anthropic, the AI safety-focused company behind the Claude model family. The AP reported that a federal judge issued an injunction blocking the Pentagon from designating Anthropic a supply chain risk, preventing the department from using that label to exclude the company from defense contracts. The AP’s reporting did not identify the specific judge or court, and those details have not appeared in other publicly available accounts reviewed for this article.
Even without those specifics, the ruling carries broader implications. It signals that AI companies shut out of Pentagon programs have meaningful legal recourse and that courts are willing to scrutinize how the Defense Department vets commercial AI vendors. For the eight companies that did secure GenAI.mil agreements, the precedent is a reminder that their positions could face legal challenge if future security reviews raise concerns. For Anthropic and other excluded firms, it opens a path to contest adverse designations.
Classified or not quite?
One of the most important unresolved questions is what classification level GenAI.mil actually serves today versus what it aspires to serve in the future.
The Pentagon’s announcement frames the agreements around classified networks, implying that at least some models will be accessible from systems handling Secret or Top Secret data. Google’s press release, by contrast, describes the initial deployment in terms of IL5 authorization, which falls below the threshold for fully classified environments. (The Defense Department uses a separate Impact Level 6 designation for Secret-level cloud workloads.)
These descriptions are not necessarily contradictory. GenAI.mil could be designed as a tiered platform: starting with IL5 environments for experimentation and lower-risk tasks, then extending into classified domains as security and oversight mechanisms mature. But no official statement has explicitly confirmed that roadmap, and the distinction matters. Security architecture, vendor access protocols, and oversight requirements differ sharply between controlled unclassified and fully classified environments.
What the Pentagon has not said
Several critical details remain absent from public documentation:
- Mission assignments: No public record specifies which company’s models will support which military functions, whether that is translation, imagery analysis, logistics forecasting, or intelligence summarization.
- Redundancy and competition: It is unclear whether multiple vendors will provide overlapping capabilities for the same tasks or whether certain mission areas will depend on a single provider.
- Human oversight rules: The Pentagon has not published guardrails governing how human operators will supervise AI outputs in time-sensitive or combat-adjacent scenarios.
- Timelines and benchmarks: No implementation schedule, performance standards, or incident-reporting process has appeared in any available public record.
Without these details, it is difficult to assess how quickly GenAI.mil will move from experimentation to day-to-day operational use, or how resilient the system would be if a vendor’s access were suspended or a model underperformed.
Commercial AI inside classified networks: what the vendor dependencies mean
The GenAI.mil agreements represent a notable shift in the relationship between Silicon Valley and the national security establishment. For years, that relationship was defined by tension: Google employees protested Project Maven in 2018, and several major tech firms kept the military at arm’s length. Now, Google, OpenAI, Microsoft, Amazon, and others are competing for positions inside the Pentagon’s most sensitive systems.
The arrangement also concentrates significant power in a small group of commercial vendors whose models were not originally built for classified environments. If GenAI.mil becomes central to how the military processes intelligence and plans operations, the Defense Department’s dependence on these companies will deepen, raising questions about supply chain risk, intellectual property, and what happens when a vendor’s commercial priorities diverge from national security needs.
The most consequential design choices are almost certainly being made on networks the public will never see. But the decisions about which companies get access, what classification levels they operate at, and how much autonomy AI tools are given in military operations will shape U.S. defense strategy for years to come. The Pentagon has opened the door. What it builds behind it is the question that matters now.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
