The Pentagon announced on Friday that it had reached deals with some of the technology industry’s biggest companies in an effort to expand the military’s artificial intelligence capabilities and increase the number of firms authorized to be on classified networks.
The companies, according to the Defense Department, agreed to allow the Pentagon to employ their technology for “any lawful use,” a standard resisted by Anthropic, which was initially the only artificial intelligence model available on classified markets.
The Pentagon had previously confirmed deals with Elon Musk’s xAI, OpenAI and Google. In addition, the Pentagon said it had reached deals with Amazon Web Services, Microsoft, Nvidia and Reflection AI, a start-up.
“These agreements accelerate the transformation toward establishing the United States military as an A.I.-first fighting force,” the Pentagon said in a statement.
Still, the announcement with the cloud computing companies is not as significant as the earlier agreements with OpenAI that allow ChatGPT and Google to use Gemini on the classified networks. The cloud companies host the models, but the agreements with the companies that create the models will determine precisely how the Pentagon uses artificial intelligence to fight wars.
In its announcement, the Pentagon did not specify how it would use the new A.I. tools but said the agreement would help service members make faster and better decisions. Some potential use cases include generating target lists for strike consideration during a conflict, such as in Iran, or analyzing vast tranches of data to glean intelligence insights.
“Access to a diverse suite of A.I. capabilities from across the resilient American technology stack will give war fighters the tools they need to act with confidence and safeguard the nation against any threat,” the Pentagon said.
Artificial intelligence has helped the military and spy agencies sift through massive amounts of intelligence. It has been particularly valuable for finding overlooked communications intercepts. The large language models are also being integrated into systems like Maven that help military personnel choose battlefield targets — though the Pentagon has not explained precisely how artificial intelligence has been used to find targets in Iran and elsewhere.
Defense Department officials hope the new deals will push Anthropic to drop its reservations about the military’s broad “any lawful use” standard.
Anthropic did not respond to a request for comment.
President Trump has ordered the government to cut ties with Anthropic, but for now the company’s technology remains on classified networks and intelligence analysts still depend on the firm’s models. While the Pentagon wants to quickly move to OpenAI’s ChatGPT and Google’s Gemini, there have been growing pains and technical problems.
Anthropic and the Pentagon are currently in federal litigation over the Defense Department’s decision to label the company a supply chain risk, a novel use of the government’s power to raise concerns about how corporations build their products.
White House officials, impressed and worried about the power of Anthropic’s newest model, Mythos, have been pushing for a compromise that would end the company’s feud with the Pentagon, or at least allow other parts of the government to work with the firm.
While Mr. Trump ordered the government to stop using Anthropic by the middle of the year, the success of the new model means that decision is all but certain to be rolled back. The military and intelligence agencies are continuing to use Anthropic’s older models for day-to-day work and are testing the capabilities of the new model to find cybersecurity exploits.
The deal with the new companies was reported earlier by Bloomberg.
A Pentagon official said the new agreements would help prevent “vendor lock” and ensure that the military would not have to depend on any one company. The military also wants firms to agree to a single standard, and has been loath to give firms contractual guarantees about how their models will be used.
Anthropic and the Pentagon have been locked in a debate over whether the company’s Claude model could be used to pilot autonomous drones or work on domestic surveillance. The Pentagon says it does not intend to use the model for either of those activities, but the two sides have not agreed on contractual language, or if it is even necessary.
The details of the deals with the new companies remain secret. But people briefed on the negotiations have said at least some of the companies have received safeguards that are akin to what Anthropic sought on drones and surveillance, even if the companies adhere to the overall standard the Pentagon has met.
The exact language of the safeguards could raise questions about exactly what is preventing the Pentagon and Anthropic from coming to an agreement.
The Pentagon and Anthropic have both, in a manner, benefited from the fight. Anthropic has won support in Silicon Valley for standing up to the Pentagon. And Defense Secretary Pete Hegseth has been able to use the fight to criticize Anthropic’s leadership as being ideological, a line that Mr. Trump has supported.
The Pentagon deals come as the White House is resisting a plan by Anthropic to roll out its Mythos model to dozens of additional companies, according to people familiar with the matter. The opposition stems chiefly from concerns that expanding access to Mythos — which has already been shared with select organizations in the United States and Britain — could pose significant cybersecurity risks.
Mythos is so adept at locating and exploiting computer bugs that Anthropic has not released it publicly, instead choosing to grant some security researchers early access to work on patching the vulnerabilities Mythos has identified.
Asked about Mythos discussions, a White House official said the Trump administration was seeking to find a balance between advancing innovation and ensuring security as it works with leading edge A.I. labs.
Dustin Volz contributed reporting.
https://www.nytimes.com/2026/05/01/us/politics/pentagon-ai-companies-deals.html
