Pentagon Adds Oracle to AI Deals With Amazon Web Services, Google, Microsoft, OpenAI, Nvidia, SpaceX

The U.S. Department of Defense has moved to broaden the set of companies whose artificial intelligence can be deployed on its classified networks, announcing agreements with eight technology firms and later adding Oracle to the list.

Breaking Defense reported that an initial press release listed seven firms—"Amazon Web Services, Google, Microsoft, OpenAI, SpaceX, NVIDIA and Reflection"—before the Pentagon CTO’s office posted on X that "Oracle has officially agreed to join the list of AI companies deploying frontier capabilities on the Department’s classified networks."

AI Insider

Decrypt likewise said the Pentagon entered agreements with eight technology firms, certifying "SpaceX, OpenAI, Google, Nvidia, Reflection, Microsoft, Oracle, and Amazon Web Services" to deploy advanced AI on classified military networks.

The agreements place the AI systems on networks classified as Impact Level 6 and Impact Level 7, with Impact Level 6 described as handling "secret data" and Impact Level 7 described as a semi-official term for the most highly classified systems.

The Pentagon framed the deals as accelerating a shift toward an "AI-first fighting force," saying the agreements will "strengthen our warfighters' ability to maintain decision superiority across all domains of warfare."

The Pentagon also said the agreements build on existing federal investment and that the announcement did not disclose the value of the contracts, while Decrypt reported that the Pentagon’s internal AI platform shows rapid early adoption across the department.

The Pentagon’s decision to approve additional AI providers for classified environments comes as it tries to avoid reliance on a single partner and as it builds on earlier steps to bring commercial AI into the Defense Department.

Breaking Defense reported that the effort builds on the standup of the secure but unclassified GenAI.mil platform in December, and it described the new agreements as integrating frontier AI capabilities into Impact Level 6 and Impact Level 7 network environments to "streamline data synthesis, elevate situational understanding, and augment warfighter decision-making in complex operational environments."

Breaking Defense

In the same report, Pentagon CTO Emil Michael said the department learned it was "irresponsible to be reliant on any one partner" and that "one partner didn’t really want to work with us in the way we wanted to work with them," a reference to Anthropic’s dispute with the administration.

Breaking Defense also said Anthropic was notably absent from the list of firms, even though "Claude AI was already in use on classified networks as part of Palantir’s Maven toolkit," and it tied that absence to lawsuits and the administration’s attempt to ban Anthropic from government work.

Decrypt and TechCrunch both described the underlying disagreement as one over usage terms and guardrails, with TechCrunch stating the Pentagon wanted "unrestricted use of Anthropic’s AI tools" while Anthropic insisted on guardrails to prevent domestic mass surveillance and autonomous weapons.

The Hill reported that Anthropic had been blacklisted about two months earlier over a disagreement on safety guardrails, and it said Anthropic was concerned its AI would be used for domestic surveillance or autonomous weapons without human oversight while the Pentagon insisted on using the technology for "any lawful purpose."

In parallel, the Pentagon’s broader push is reflected in its internal platform metrics, with Decrypt reporting that GenAI.mil launched in December and that "more than 1.3 million personnel have used it to generate tens of millions of prompts and deploy hundreds of thousands of AI agents in five months."

The Pentagon’s announcement included both a policy rationale and responses from companies and outside advocates, with multiple outlets quoting the same core language about decision superiority.

“UPDATED 5/1/26 at 1:05 pm and 4:05 pm: This story was updated with further details from DoD and to reflect a later announcement that Oracle had been added to the list of companies to deploy on classified networks”

Breaking Defense

Breaking Defense quoted the Pentagon’s initial statement that "These agreements accelerate the transformation toward establishing the United States military as an AI-first fighting force and will strengthen our warfighters' ability to maintain decision superiority across all domains of warfare," and it also included the Pentagon CTO’s remarks that the department went out to ensure "multiple different providers — both open source" and proprietary model companies and infrastructure companies like Microsoft and AWS.

Decrypt quoted an OpenAI spokesperson saying, "As we said when we first announced our agreement several months ago, we believe the people defending the United States should have the best tools in the world," and it quoted an AWS spokesperson, Tim Barrett, saying, "For more than a decade, AWS has been committed to supporting our nation's military and ensuring that our warfighters and defense partners have access to the best technology at the best value."

Decrypt also included a direct warning from Greg Nojeim, director of the Center for Democracy and Technology’s Security and Surveillance Project, who told the outlet, "The DoD announcement that it has agreed to deploy AI on classified networks raises more questions than it answers."

Nojeim pressed on oversight and lethal outcomes, saying, "How will DoD use the AI that it deploys, and how will it ensure that such use does not result in errant decisions with lethal impact?" and he added, "Will it use AI to further supercharge surveillance, including surveillance of Americans?"

TechCrunch similarly quoted the Pentagon’s statement that "The Department will continue to build an architecture that prevents AI vendor lock-in and ensures long-term flexibility for the Joint Force," and it said the department’s goal was to give warfighters a "diverse suite of AI capabilities" to act "with confidence."

In Gizmodo’s account, AWS spokesperson Tim Barrett repeated the company’s commitment and said, "We look forward to continuing to support the Department of War’s modernization efforts, building AI solutions that help them accomplish their critical missions," while Gizmodo also described Anthropic’s earlier refusal to accept language that would allow "any lawful purpose."

While the Pentagon’s core announcement is consistent across outlets, the reporting diverges in emphasis—particularly around Anthropic, oversight concerns, and the number and identity of firms.

Breaking Defense described the initial list of seven firms and then the later addition of Oracle, while also stressing that Anthropic was absent and that Claude AI was already in use via Palantir’s Maven toolkit, and it included a quote from Emil Michael on CNBC that criticized reliance on any one partner.

Coinpaper

Decrypt framed the deals as part of a broader push to integrate AI across military operations and highlighted GenAI.mil adoption, quoting the Pentagon’s view that the system will support "data analysis, situational awareness, and decision-making" and is designed to use multiple AI providers rather than rely on a single vendor.

TechCrunch emphasized the vendor-diversification rationale and said the Pentagon wanted to avoid vendor lock-in, quoting the Pentagon’s line about "long-term flexibility for the Joint Force" and describing IL6 and IL7 as high-level security classifications requiring protection "physically, through strict access controls and audits."

The Hill, meanwhile, focused on the Anthropic dispute and the administration’s earlier blacklisting, stating that Anthropic was notably not included and that it had been designated a "supply chain risk" earlier this year.

Gizmodo framed the same expansion as "The Rest of Big Tech Piles in to Take the Pentagon Deal That Anthropic Wouldn’t," and it described the deals as coming as concerns grow around surveillance and military applications, while also quoting AWS’s Tim Barrett and describing Anthropic’s negotiations over language allowing "any lawful purpose."

Cybernews and crypto.news both described the agreements as expanding the roster of tech partners and explicitly listed the firms, but they differed in how they characterized the Pentagon’s confirmation of Google’s agreement, with crypto.news calling it "the first formal confirmation from the Pentagon" while Cybernews described the Pentagon reaching agreements with seven leading AI companies.

Beyond the classified-network approvals, the technology landscape described in the sources also includes shifting cloud access for OpenAI models and ongoing Pentagon reliance on multiple providers.

Ecosistema Startup reported that Amazon Web Services will begin selling OpenAI models to its cloud customers in April 2026, saying it will start "un día después de queMicrosoftaceptara terminar el acuerdo exclusivo de reventa" that had given Azure unique access to OpenAI technology during the first three years of generative AI.

crypto.news

That same article said the strategic agreement between AWS and OpenAI is valued at "38.000 millones de dólares," with potential expansion up to "100.000 millones" in eight years, and it asserted that founders would have "por primera vez desde 2019" a real option to access OpenAI technology without being tied exclusively to Azure.

In the Pentagon context, Decrypt said the agreements did not disclose when the AI models would be available on classified networks or how much the companies are being paid, and it quoted a Pentagon statement that the agreements build on existing federal investment.

TechCrunch similarly described the Pentagon’s classified deployment as for "lawful operational use" and said the deals come as the Defense Department accelerates diversification of AI vendors after its dispute with Anthropic over usage terms.

Looking at the operational platform, Decrypt and Tom's Hardware both tied the classified-network push to GenAI.mil, with Decrypt reporting that "more than 1.3 million personnel" used the platform and Tom's Hardware quoting the Pentagon that "Over 1.3 million Department personnel have used the platform" and that it generated "tens of millions of prompts" and "hundreds of thousands of agents" in "only five months."

The sources also show that the Pentagon’s approach is explicitly designed to avoid vendor lock-in, with TechCrunch quoting the Pentagon’s architecture language and Decrypt describing the system as designed to use multiple AI providers rather than rely on a single vendor.