On March 5, LASST filed two Freedom of Information Act requests (here and here) on behalf of investigative reporter Sam Biddle of The Intercept, seeking the agreements governing the deployment of frontier AI systems on Department of Defense networks. We requested the contracts between the Pentagon and four companies — xAI, Google, Anthropic, and OpenAI — including the now-infamous agreement OpenAI announced on February 27.
We also requested expedited processing, arguing that there is a compelling need for these records and an urgency to inform the public about them.
The Department of Defense denied that request. Its reasoning deserves scrutiny.
What We Asked For and Why
Over the past several weeks, the relationship between the U.S. military and America’s leading AI companies has undergone a dramatic and public evolution. The Pentagon designated Anthropic, an American company, as a supply chain risk to national security, a designation historically reserved for foreign adversaries. President Trump ordered federal agencies to cease using Anthropic’s technology. Hours later, OpenAI announced its own deal with the Pentagon, which its CEO, Sam Altman, later acknowledged on X was “rushed” and appeared “opportunistic and sloppy.”
These events have generated extraordinary public attention, active congressional oversight, and federal litigation. Anthropic has sued the government, alleging that the supply chain risk designation is unlawful retaliation for the company’s refusal to allow its AI to be used for mass domestic surveillance or fully autonomous weapons. The case raises questions at the intersection of national security, civil liberties, the First Amendment, and the laws of armed conflict.
But the public has yet to see any of the actual contracts. We know what Altman has posted on X. We know what Pentagon officials have said in press statements. But the binding legal text, the language that will actually govern how AI is used by the American military, remains hidden.
The Pentagon’s Denial of Expedited Processing and our Appeal
The Department denied our requests (here and here) for expedited processing of our document requests. The stated rationale was that we had “not clearly demonstrated how the information will lose its value if not processed on an expedited basis.” The request was placed in the complex processing queue, where it could languish for months or years.
This denial is difficult to reconcile with the legal standard or with reality.
Under FOIA, expedited processing must be granted when a requester who is primarily engaged in disseminating information demonstrates an “urgency to inform the public concerning actual or alleged Federal Government activity.” The Courts have held that this standard is met when the request concerns a matter of current exigency to the American public, the consequences of delay would compromise a significant recognized interest, and the request concerns federal government activity.
Every element is present here. The Pentagon’s actions against Anthropic and its simultaneous embrace of OpenAI are matters of intense and ongoing public debate. Members of Congress have publicly questioned the supply chain risk designation. Federal courts are adjudicating the legality of the government’s conduct. Media coverage has been extensive and sustained. The agreements themselves are at the heart of every one of these disputes. And not one of them has been made public.
The Department’s denial letter does not engage with these facts. It does not explain why the agreements lack time-sensitive value. It does not address the pending litigation, the congressional scrutiny, or the weeks of front-page coverage. It simply recites the legal standard and declares it unmet. That is not the reasoned analysis FOIA requires.
On March 18, we appealed the Pentagon’s denial of expedited processing (here and here). On March 19, the Pentagon responded (here and here). The Pentagon’s response includes this remarkable statement: “Due to an extremely heavy FOIA workload, we are unable to complete your appeal within the statutory time requirement.” In other words, they are going to break the law that requires them to decide the appeal quickly.
Why This Matters
OpenAI claims its agreement contains meaningful protections against domestic surveillance and autonomous weapons. Critics, including former military officials, privacy advocates, legal scholars, and even current OpenAI employees, say the disclosed language is riddled with loopholes. The whole episode raises questions about what the DoD’s contracts with the other big AI companies say.
None of these questions can be resolved without the underlying documents. As others have recently observed, the government is building a military AI governance regime through procurement contracts rather than statutes or regulations, and the public cannot evaluate that regime when the contracts are secret.
The public deserves to know the terms on which frontier AI systems are being deployed in classified military environments.
What Comes Next
We will continue to press for the timely production of these records and continue that fight in court, if necessary.
When the government deploys the most powerful AI systems ever built and punishes companies that insist on safety guardrails, the public has a right to see the terms.