'Nobody really knows:' Pentagon clash with Anthropic throws agencies into limbo |
‘Nobody really knows:’ Pentagon clash with Anthropic throws agencies into limbo
Federal agencies and their contractors have been left in limbo as the Trump administration moves to cut off Anthropic from government systems without formal orders amid a brewing legal battle with the AI company.
As agency leaders grapple with informal directives from President Trump and the Pentagon, the situation is exposing the challenges and costs of removing a major AI vendor from federal supply chains after an aggressive push to embed the technology in the first place.
Nearly three weeks have passed since Trump ordered federal agencies to “immediately cease” using Anthropic’s technology, but various federal agencies have yet to receive formal guidance other than Trump’s social media post on how to proceed, according to conversations with multiple federal technology leaders.
In turn, the response has varied across the government, with agencies like the General Services Administration and the Department of Health and Human Services abruptly removing Claude within hours of Trump’s directive. Other agencies say they are still reviewing Anthropic’s use, but the product may still be available.
Trump’s directive followed a breakdown in negotiations between the Pentagon and the AI company earlier this month over disagreements on safety guardrails. Defense Secretary Pete Hegseth separately deemed Anthropic a supply chain risk, which will be fought over in federal courts later this month.
Federal employees get few answers
For staffers at agencies that already moved to eliminate Anthropic, the transition has been confusing and abrupt, federal tech leaders told The Hill.
Anthropic was first approved for classified use in government agencies through a partnership with Palantir nearly two years ago. Since then, the Trump administration has pushed federal agencies to use AI in workflows, leading to a rapid adoption of technology, including Anthropic’s Claude models, across defense and civilian spaces.
At HHS, thousands of employees using Anthropic products had just a few hours to save their chats and coding projects, according to an agency leader.
“Staff were really upset with how quickly” the shutdown happened, the leader said, adding “there was no spin-down time.”
“People lost their chats, people lost any coding that they were doing in any projects. Are there equivalent tools that they can use? Sure, but they had been working in a secure environment,” the leader added. “It’s a loss of a lot of work…it was a waste of government resources.”
AI leaders across HHS received notice about the pending elimination less than an hour after Trump posted his directive on Truth Social, according to a screenshot obtained by The Hill. That notice, sent by HHS Deputy Chief AI officer Arma Sharma, said Claude Enterprise would be disabled “in alignment” with Trump’s directive.
Days later, another message clarified enterprise access to Claude was “temporarily disabled,” at the agency, but HHS’s office of the chief AI officer was “awaiting more detailed federal guidance regarding the future use of applications and systems that leverage Claude or other Anthropic technologies.” Staff were told more direction would come pending “more definitive guidance.”
HHS confirmed ChatGPT Enterprise and Google Gemini remain available for staff.
Reports circulated soon after that the White House is floating an executive order to eliminate Anthropic’s AI from the government, though this hasn’t come to fruition. The General Services Administration, the agency responsible for most federal technology procurement, is also proposing a clause to existing and new GSA schedule contracts that would confirm the government’s right to use an AI system “as necessary for any lawful Government purpose.”
The clause would apply to the AI firms, as well as subcontractors or vendors, and is similar to the demands of the Pentagon, which maintains it should be able to use AI technologies for “any lawful purpose” in the military.
GSA also removed Anthropic from its governmentwide AI testing tool, USAi and terminated its OneGov deal with the firm, which offered agencies the chance to use the company’s tech at near-zero costs.
At another civilian agency, one AI advisor told The Hill there was “a tremendous lack of information,” and “nobody has clear answers” even as agency leaders told workers to stop using Anthropic’s technology.
The advisor, who spoke on the condition of anonymity to speak freely, compared the confusion to the chaotic takeover of Trump’s so-called Department of Government Efficiency, which sparked more questions than answers for federal workers last year.
Civilian AI leaders, according to the advisor, are still unsure of whether the order applies to all of the federal government, including contractors who may use Anthropic in their own workflows but not directly in their work for agencies.
“It’s a lot of complicated questions that nobody really knows the answer to,” the leader said, adding their agency told them to “stop using” Anthropic products and that they will “get back” to them with more details.
One federal technology leader familiar with procurement suggested “some political [appointees] seem to be proactively ordering staff based on social media, but that’s up to them.”
Some agencies have yet to clarify
Meanwhile, some agencies are holding their breath, at least publicly. It is unclear how critical missions, such as nuclear weapons research, will be impacted by the situation.
The Department of Energy’s National Nuclear Administration and national labs have partnerships with Anthropic to work on nuclear weapon risk research and assist scientists, respectively.
When asked how the agency plans to proceed, a DOE spokesperson said Tuesday the agency is “reviewing all existing contracts and uses of Anthropic technology,” and is “committed to ensuring” the technology it uses “serves the public interest” and “protects America’s energy and national security.”
Anthropic, which filed a suit against the Trump administration over the supply chain risk designation, argues the determination should only impact Claude customers on contracts with the Department of Defense, not all Claude customers who have the contracts.
This differential may be determined in the court case, and Anthropic’s lawyers noted in their complaint last week that agencies already took action despite the uncertainties.
“Throughout, the federal government has never once expressed concerns about Anthropic’s security or Claude’s competencies,” attorneys wrote, pointing to Anthropic’s FedRAMP High authorization through Palantir.
The Department of Treasury and the Secret Service also stopped the use of Claude, FedScoop reported last week.
Other agencies including the Department of Veterans Affairs and the Office of Personnel Management, which listed Anthropic products on their 2025 AI use case inventories, did not respond by publication time on their plans, while NASA referred The Hill to the Justice Department. OPM’s updated inventory, posted last week, shows Anthropic was removed from an earlier version.
Technology leaders both in and outside of government are also sounding the alarm on the costs of this termination, and what it means for the taxpayer at the end of the day.
Franklin Turner, the co-chair of McCarter & English’s Government Contracts practice group, predicted there will be a cost impact to the government.
Should a subcontractor say they are using Anthropic, agencies “would have to terminate that subcontract” and “go out and find a new one,” Turner told The Hill.
“That carries with it a cost and that’s a cost that wasn’t foreseen at the time you prepared and submitted your bid,” he added.
Chris Griesedieck, a government contracts attorney at Venable LLC, echoed this sentiment, telling The Hill contractors may also be willing to make the modification to comply, but “[the contractor] reserves the right to an equitable adjustment if this is going to cost me a bunch of extra money.”
The HHS leader, also speaking on the condition of anonymity, added the agency’s abrupt removal of Anthropic “wasted taxpayer dollars.”
“Phasing out of it would have been annoying, but it wasn’t, it was shut down immediately and everybody’s work was lost,” the leader said, adding, “Agencies who built programmatic systems on it, they’re gonna have a ton of work.”
Copyright 2026 Nexstar Media Inc. All rights reserved. This material may not be published, broadcast, rewritten, or redistributed.
Judge skeptical over Trump ballroom project amid new bid to halt it
Judge permanently blocks Ten Commandments displays at several Arkansas school ...
Trump allies plan Senate floor takeover to pass SAVE America Act
Former Trump appointee: MAGA movement is ‘dead’
Republicans collide with Trump over no-excuse absentee voting, SAVE Act
Joe Kent resignation over Iran war reveals an internal MAGA Trump division
Army general left classified maps on train, concussed after ...
Johnson says Democrats’ pitch to fund DHS without ICE, CBP would ‘defund ...
Biss wins Democratic primary in Illinois House race defined by Israel support
Senate votes to begin marathon debate on SAVE America Act
Comer formally subpoenas Pam Bondi over Epstein investigation
Stratton on track to succeed Durbin: Key takeaways from Illinois primary ...
Sununu on TSA workers, DHS shutdown: ‘I don’t think there’s ...
GOP lawmaker argues Marines on Kharg Island would not be 'boots on the ground'
Crenshaw on primary defeat: GOP voters bought into ‘misinformation’
Trump administration expands Medicaid fraud crackdown to Florida
Inside Noem’s tense relationship with controversial DHS inspector general
Live updates: Stratton advances in Democratic Senate primary; Biss, ...