The US Department of Defense no longer uses the artificial intelligence tool Cloud, as stated on October 5 by a Pentagon official, who indicated that the department “Enthropic has stopped using the product”.
Defense Secretary, Pete Hegsethhad set the end of August as a deadline and no one explained why it took more than a month to complete the withdrawal
Until the end of September, Claude was used For research, analysis and intelligence gathering, as well as for military operations against Iran, according to former Defense Department officials and contractors who worked with the Pentagon on artificial intelligence projects, speaking on condition of anonymity because they were not authorized to speak to the press. For these sources, Claude was a tool Essential for processing large amounts of dataAnd Anthropic declined to comment on the matter.
The commercial relationship began in 2024, when Anthropic, Palantir and Amazon Web Services announced an alliance in November. Bring Claude to the intelligence and defense agencies In a certified environment for classified information, called Level IL6. Palantir was then proud to be the first partner to bring cloud models to a hierarchical environment. In early 2026, the Pentagon asked Anthropic to remove Claude’s security measures and give the military unrestricted access.
However, the company declined on two specific points and in its statement on February 27, Exceptions are requested for both Mass surveillance of American citizens For fully autonomous weapons. that day, Pete Hegseth Announced that the company will be nominated “Supply Chain Risk” of national security (a label usually reserved for suppliers from countries considered hostile), and Donald Trump ordered federal agencies to stop using their products.
The six-month deadline began the day before the U.S. and Israel attacked Iran, and Anthropic, which had already called the move “unprecedented,” sued the government on March 9.
Introduction to Claude in Maven
Claude was part Maven Smart SystemThe platform is operated by Palantir (founded by billionaire Trump ally Peter Thiel) and used by the Pentagon as the primary system for organizing intelligence and other data.
Analysts are said to have flocked to Maven Satellite images and drone recordings And then they pass it to Claude (and other large language models) for tasks like identifying potential military targets. A person who was demoed by Palantir earlier this year said it was often used to create one-page reports that went up the chain of command.
Some US allies also use Maven To share intelligence with Washington. Moreover, the role of the model is brief: an analysis of Claude at CENTCOM (the Central Command conducting operations in the Middle East), published on March 30, describes him as a part that integrates modules and models of the system rather than elements that propose objectives.
This analysis also lists that Department of Defense policy requires individuals to select targetsAnd that on March 9 the Under Secretary of Defense Steve Feinberg He directed Maven to the Pentagon’s Office of Central Information and AI.
That explains why the merger wasn’t immediately withdrawn. “The fact that the Pentagon has gone so far as to eliminate all use of clades in the system indicates that these things are not just ‘plug and play'”He said Lauren Kahn, an analyst at Georgetown University’s Center for Security and Emerging Technologies and former defense officer. And he added: “Once they get together, they can be hard to eradicate.”.
Pentagon personnel continue to use Claude with the Mythos model, “During the entire duration of the dispute”According to source one. And when the order came, “There were concerns about moving from Entropic to OpenAI” And in other tools, recalled a former high-ranking official.
The Mythos detail is interesting because it was not an accessible model to anyone. Anthropologie presented the Cloud Mythos preview on April 7 within the GlassWing project, reserved for a closed group of partners (Microsoft, Apple, Google, Cisco or CrowdStrike, among others) and more than 40 organizations that maintain critical software, due to the model’s ability to find and exploit vulnerabilities. The company itself cautioned that it has no plans to open that version to the general public.
Relations with the White House appear to be different, as between late September and early October, President Trump met twice. Dario AmodeiCEO of Anthropic. Amodi told the press that he was concerned about the risks of artificial intelligence: “How to address these risks is still being discussed.” however, “If we work with the President, we can win safely”He added, and Trump described it himself “bright” And “excellent”.
OpenAI has established very similar limits
The Pentagon is closed Also deals with Google, xAI and OpenAIAnd two people familiar with the Defense Department’s use of artificial intelligence said OpenAI tools have spread to some departments in recent months.
On the night of February 27, Sam Altman The agreement with the Pentagon announced and assured that the department has adopted two of the company’s policies: domestic mass surveillance and humanitarian responsibility prohibitions on the use of force, including autonomous weapons.
On paper, these limits are very similar to those landing ethnographers on the blacklist. However, Altman stated that The difference was that they were already in Law and Defense Department policyAnd that OpenAI has incorporated them into the agreement with Technical Security. Altman himself admits that domestic mass surveillance is already illegal.
Two courts, two answers
The two designations are based on different values. The first is based on Section 3252 of Title 10 of the US Federal Code and Federal judge Rita LynnIn California, August 27 outlawed it: it considered it retaliation that violated freedom of expression, an arbitrary decision without due process, and wrote that national security was not invoked. “a blank check” To punish critics of the government. That judgment remains in effect.
A month later, on September 25, the District of Columbia Court of Appeals The other supported the designation by a two-to-one vote, based on the Federal Procurement Supply Chain Security Act (FASCSA). The majority considered that Claude’s restrictions and lack of closure could make it unreliable for military operations, with Justice Karen Lecraft voting against Henderson. Anthropic recalled that another court had already invalidated the parallel designation and was considering requesting a new review
If you use Claude?
If someone uses the cloud at home or in their office, nothing changes: Anthropologist clarified its position on February 27 and explained that access for individuals and companies is not affected by the conflict.
For companies that work with the Pentagon, however, The situation is more confusing. According to an analysis of the September ruling, the department asked some contractors to remove anthropomorphic products from certain systems and deliverables, and even suggested banning them on systems not part of the contract, which could go beyond FASCSA’s permits. Even the Federal Contracting Registry (SAM) has yet to publish a formal exclusion order.
As the FASCSA designation remains in effect, Claude remains outside of Maven, and Anthropic has not specified which review it will request.
Picture | With image GPT and Instagram editing
