“It’s impossible to assess any safeguards that may exist in your company’s agreement with DoD without seeing the full contract,” Sen. Elizabeth Warren wrote.
A Senate lawmaker is demanding that the Defense Department and seven technology companies deploying their artificial intelligence capabilities on the department’s classified networks disclose the terms of their agreements in order to better understand what guardrails DoD is putting in place around AI use in the military.
In a letter to Defense Secretary Pete Hegseth, Sen. Elizabeth Warren (D-Mass.) said the department “appears to have forged ahead in expanding AI use throughout the military, while releasing no meaningful information about contractual guardrails to prevent the misuse of these tools or communicating to Congress and the public how these tools are used at all.”
In May, the Defense Department struck agreements with SpaceX, OpenAI, Google, NVIDIA, Microsoft, Amazon Web Services, Reflection and Oracle to integrate their AI capabilities into the department’s Impact Level 6 and Impact Level 7 network environments. IL6 is used for the storage and processing of information classified up to the secret level, while IL7 supports highly restricted data.
In a separate letter to the seven companies, Warren raised concerns that they may have entered into contracts that would give the Defense Department and Trump administration officials “free rein to engage in domestic surveillance – including spying on U.S. citizens exercising their legal rights — or build autonomous weapon systems that have enormous power to make decisions about targeting without human intervention.”
“Ultimately, it is impossible to assess any safeguards and prohibitions that may exist in your company’s agreement with DoD without seeing the full contract, which neither DoD nor your company have made available,” Warren said.
“Despite prior Congressional requests, DoD has provided no details of the breadth or substance of the contracts beyond that they allow the technology to be used on classified systems for all ‘lawful operational use,’” she added.
The Massachusetts senator argued that even if the contracts include safeguards — like bans on domestic surveillance or limits on autonomous weapons use — the language may be too broad or vague to actually restrict DoD from conducting mass surveillance or using lethal autonomous weapons with limited human oversight.
Some experts pointed out that OpenAI’s contract with DoD, the snippets of which the company released in February, does little to limit how the Defense Department could use AI, arguing that references to “all lawful purposes” and “applicable law” in the contract “cannot be relied on to constrain the government’s actions” without clearly defined contractual limitations. Another expert said the terms related to mass surveillance are so broad the government can “drive a truck through [them].”
“As long as the Department can make a plausible case that its conduct is not ‘unlawful,’ it appears to be free to use your technology in a manner that harms civilian populations and endangers civil liberties,” Warren said.
The lawmaker also asked about the department’s agreement with Reflection AI, a startup that has yet to publicly release an AI model. The company has financial backing from venture capital firm 1789 Capital, where President Donald Trump’s son, Donald Trump Jr., is a partner. Warren asked Hegseth about the criteria the Defense Department used to award the contract to Reflection and whether any department officials communicated with Trump Jr. or his representatives about the contract.
When the Defense Department announced it struck deals with the tech companies, DoD officials said the agreements would “prevent AI vendor lock and ensure long-term flexibility for the Joint Force.”
The New York Times, however, previously reported that the Defense Department might have used negotiations with the companies to push Anthropic to drop its safety concerns about military AI usage. The Defense Department designated Anthropic a “supply chain risk” after the company rejected the Pentagon’s demand to remove two narrow contractual limitations on how DoD can use its AI models.
“These agreements come in the midst of DoD’s ongoing litigation against Anthropic after the company objected to their models being used for all lawful use cases, which would have removed preexisting safeguards preventing its technology from being used to develop fully autonomous weapons and engage in mass domestic surveillance,” Warren said.
More broadly, Democratic lawmakers have been pushing to put guardrails around the Pentagon’s use of artificial intelligence in military operations. Sen. Elissa Slotkin (D-Mich.) introduced a bill that would prohibit the Defense Department from using autonomous weapons to kill a target without human authorization and ban the department from using AI for mass surveillance and launching nuclear weapons. In June, Sen. Kirsten Gillibrand (D-N.Y.) introduced a similar bill. The Senate Armed Services Committee added the language restricting the military use of AI to its version of the 2027 defense policy bill.
Warren asked the Defense Department and the companies to provide responses in an unclassified form by July 20.
If you would like to contact this reporter about recent changes in the federal government, please email anastasia.obis@federalnewsnetwork.com or reach out on Signal at (301) 830-2747.
Copyright
© 2026 Federal News Network. All rights reserved. This website is not intended for users located within the European Economic Area.
