The Pentagon has signed agreements with seven commercial technology companies to deploy artificial intelligence tools on classified military networks, placing firms better known for consumer products and commercial cloud services at the core of some of the most sensitive computing environments in the U.S. government.
The companies are SpaceX, OpenAI, Nvidia, Google, Microsoft, Amazon Web Services, and Reflection. The deals, confirmed in a May 2026 release from the Department of Defense (which now operates under the rebranded war.gov domain), mark a deliberate turn away from legacy defense contractors and toward Silicon Valley for military AI capability.
The one deal with a price tag
Only one of the agreements has a publicly confirmed dollar value. OpenAI Public Sector LLC received a fixed-amount prototype Other Transaction Agreement, contract number HQ0883-25-9-0012, valued at $200 million, according to official Pentagon contracting records. OTAs allow the Defense Department to sidestep traditional procurement timelines, giving non-traditional contractors a faster route into military programs.
For the remaining six companies, no contract values, scope documents, or itemized award notices have been published. The public record consists of the departmental announcement and wire reporting, not the kind of granular contracting data that would reveal what each firm is expected to deliver or how much it will be paid.
Who already had a foot in the door
Three of the seven companies already operate inside the Pentagon’s classified cloud architecture. Amazon Web Services, Google, and Microsoft hold positions under the Joint Warfighting Cloud Capability program, a multi-vendor contract awarded in late 2022 that provides cloud services at every classification level, from unclassified to top secret. Oracle also holds a JWCC position, a detail that becomes relevant below.
That existing infrastructure gives AWS, Google, and Microsoft a structural advantage. Running frontier AI models on classified networks requires computing environments that meet strict federal security standards. Companies that have already cleared those hurdles do not need to build new accredited facilities from scratch. SpaceX, Nvidia, OpenAI, and Reflection, none of which are JWCC awardees, likely depend on different contract vehicles or partnerships with existing cloud providers to access those environments.
The Oracle question
The exact vendor count depends on the source. The Associated Press reported seven companies. The Defense Department release, however, lists eight, adding Oracle alongside the other seven. Whether Oracle holds a distinct AI agreement or whether its existing JWCC role was grouped in with the new deals has not been publicly clarified. The headline of this article reflects the AP’s count of seven, but readers should be aware that the government’s own list is longer.
What Reflection actually is
Most of the companies on this list need no introduction. Reflection is the exception. Reflection AI is a defense-focused artificial intelligence startup that has built tools for military planning and intelligence analysis. Its inclusion alongside household names like Google and Amazon signals that the Pentagon is not limiting these agreements to the largest firms; it is also pulling in smaller companies with specialized national-security expertise.
What the Pentagon says it wants
The Defense Department has framed the effort as a modernization push to bring commercial AI into operational military workflows, not just back-office analytics or experimental sandboxes. The AP reported that officials described the goal as augmenting decision-making in complex operational environments.
That language is broad by design. The department has not disclosed which AI models will be deployed, what tasks they will perform, or what classification level of data they will process. The release references Impact Level environments but does not specify whether the work targets Impact Level 5 (national security data) or Impact Level 6 (secret-level workloads), a distinction that carries different security requirements, audit obligations, and data-handling restrictions.
None of the seven companies have issued detailed public statements about the specific terms of their classified-network agreements, the models they plan to deploy, or their delivery timelines.
The conflict-of-interest question no one has answered
SpaceX’s inclusion raises a question the Pentagon has not publicly addressed. Elon Musk, SpaceX’s CEO, has held an advisory role in the current administration through the Department of Government Efficiency initiative. Whether Musk had any involvement in shaping the procurement process that led to SpaceX’s selection, or whether formal recusal procedures were followed, is not documented in any publicly available record. The absence of disclosure on this point is notable given the scale of the agreements and the sensitivity of the networks involved.
Integration risks across a fragmented vendor base
Deploying AI from seven different commercial vendors on classified networks creates practical challenges the Pentagon has not publicly detailed. Each company uses different model architectures, different training pipelines, and different security baselines. Making those systems work together requires shared interface standards, common logging and audit mechanisms, and clear rules governing how one model’s output can feed into another’s input.
Governance is equally murky. If a model deployed under one of these agreements contributes to a flawed intelligence assessment or an operational error, the public record does not explain how responsibility would be divided among the vendor, the cloud provider hosting the workload, and the military command that acted on the output. Traditional weapons programs have decades of testing doctrine and legal frameworks for assigning accountability when systems fail. These AI agreements, as described so far, do not come with comparable transparency.
Security concerns go beyond software vulnerabilities. Running commercial AI on classified data raises questions about how training, fine-tuning, and model updates will be managed. The available documents do not specify whether models will be trained exclusively on government-provided datasets, how access to training data will be controlled, or what safeguards will prevent classified information from leaking into future commercial products.
Where the traditional defense giants fit
Notably absent from the vendor list are the companies that have dominated Pentagon contracting for decades: Lockheed Martin, Raytheon (now part of RTX), Northrop Grumman, and General Dynamics. Also missing are Palantir and Anduril, two firms that have built their businesses specifically around defense and intelligence AI applications.
Their absence suggests the Pentagon sees the fastest advances in large language models, computer vision, and autonomous reasoning happening in the commercial tech sector rather than inside defense-specific R&D programs. The prototype OTA mechanism used for the OpenAI deal reinforces that reading, since OTAs are specifically designed to bring non-traditional contractors into defense work without the overhead of standard procurement.
What to watch next
The clearest picture right now is of a defense establishment placing a large bet on commercial AI while asking the public to accept much of the implementation on faith. The OpenAI prototype OTA shows that significant money is already committed. JWCC demonstrates that the classified cloud plumbing is in place. And the Defense Department release confirms that a cluster of high-profile technology companies has agreed to bring their models into secure military environments.
What remains opaque: the precise scope of each deal, the safeguards governing deployment, the metrics that will determine whether these systems actually improve military decision-making, and whether the procurement process that selected these vendors can withstand scrutiny on conflicts of interest. Until task orders, performance milestones, or independent audits enter the public record, these agreements remain statements of intent backed by one confirmed $200 million contract and a great deal of institutional ambition.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
