Skip to content
Today
Sections

Build Independently confirmed2 publishers 2 min read Published Updated

Pentagon's removal of Claude ran past Hegseth's six-month deadline

Pentagon says it has stopped using Anthropic's Claude after a phaseout that, per the BBC, overran Pete Hegseth's six-month deadline. Because Claude was built into Palantir's Maven, taking it out was harder than cutting off access to a chatbot.

The Engineer · Build desk

Send a correction

What happened

  • Sources including former defense officials and contractors told the BBC that the military was still using Claude last week, including in operations involving Iran.
  • The dispute began when Anthropic CEO Dario Amodei refused to remove safeguards against mass domestic surveillance and fully autonomous weapons, while backing other lawful national-security uses.
  • Hegseth announced a supply-chain-risk designation against Anthropic on February 27th, and the department made it formal on March 3rd.
  • The D.C. Circuit turned down Anthropic's challenge to the designation on September 25th, with Judge Karen LeCraft Henderson dissenting.

Why it matters

  • What doesn't add up contradiction Until the department dates its final cutoff, its 'ceased' statement and the BBC's sourcing leave the real end of the phaseout open.
  • What it sets up precedent Model makers that keep use limits in defense contracts now face a supply-chain-risk label that the D.C. Circuit has already upheld once.
  • What it costs cost With replacement contracts already awarded, the cost of dropping Claude fell on the teams migrating and modifying the workflows that used it.
  • What it limits constraint Embedding one model deep in a shared platform like Maven turns any later policy-driven switch into a software migration, slower than revoking a chatbot account.

The contract side of the switch was already in place. In July 2025 the Pentagon had given contracts to Anthropic, Google, OpenAI and xAI, so swapping out Claude did not by itself call for a new contract [17]. DefenseScoop reported that as of October 5th Anthropic models were out of the department's systems, and that it was using Google, xAI and OpenAI models or moving over to them [12]. Those are the other three companies on the July awards [21]. The BBC reports that OpenAI tools have gained wider adoption in some military departments [11].

The software side took longer. Palantir runs Maven, a system that organizes intelligence and other data for the Pentagon [7]. Several people familiar with how the department uses AI told the BBC that Claude had become an important tool inside it for processing large volumes of information [8]. The BBC's sources said it was used for research, analysis and intelligence work [4]. The Defense Department's Chief Digital and Artificial Intelligence Office said some workflows required migration and modification [18].

In my view that migration work, more than any contract step, explains why the transition ran long [2]. Each modified workflow has to work on the new model before the old one can be switched off. Lauren Kahn, a former defense official who is now a senior research analyst at Georgetown University's Center for Security and Emerging Technology, told the BBC that because Claude was so integrated, it could not be pulled out quickly [9]. "Once they become integrated it can be painful to remove them," she said [10].

A Pentagon official said the department "has ceased the use of Anthropic products" [1]. The BBC's sourcing on recent use [3] and DefenseScoop's October 5th date [12] describe different points in one transition. Neither the statement nor the reporting says when final access ended, whether every contractor environment finished, or how many users and systems were affected [20][19].

The dispute was about a contract term from the start. In a February 26th statement, Amodei wrote that Anthropic "cannot in good conscience accede" to the Pentagon's request for an "any lawful use" term [6]. The D.C. Circuit majority found that Anthropic's ability and willingness to limit how Claude is used was something the department could treat as a supply-chain risk [15]. It said the department acted over a contract term it considered essential to national security [15]. According to the runtimewire account of the BBC reporting, the ruling did not settle how quickly a deployed model can be removed once it is built into operational software [16].

What to watch

  • A dated Pentagon statement on when final Claude access was removed, and whether contractor environments such as Maven are covered by it.
  • Whether Anthropic seeks further review of the D.C. Circuit ruling that drew Judge Henderson's dissent.
  • Any CDAO or Palantir detail on how many Maven workflows had to be migrated to Google, xAI or OpenAI models.

Clarity's read

What the record supports and how the coverage leans. The claims behind it follow.

Reality

Evidence58
Adoption62
Hype gap+15
Incentives60
Confidence55
Why these scores

Claim ledger

Ranked by verification strength, evidence, and original report placement.

  1. [1]

    A Pentagon official said the department "has ceased the use of Anthropic products."

    Reported Supported Source: Pentagon official, via BBCView cited source
  2. [2]

    The Pentagon's claimed end of Claude use came at the end of a transition that ran beyond the six-month phaseout period set by Defense Secretary Pete Hegseth, according to the BBC.

    Reported Supported Source: BBC NewsView cited source
  3. [3]

    Sources who spoke to the BBC, including former defense officials and contractors, said Claude was still in use as recently as last week, including in military operations involving Iran.

    Reported Supported Source: BBC sourcesView cited source
  4. [4]

    BBC sources said Claude had been used for research, analysis and intelligence work, and was embedded in Palantir's Maven Smart System.

    Reported Supported Source: BBC sourcesView cited source
  5. [5]

    Anthropic CEO Dario Amodei refused to remove safeguards against mass domestic surveillance and fully autonomous weapons, saying Anthropic's limits applied to those two areas while it supported other lawful national-security uses.

    Reported Supported Source: Dario Amodei, via runtimewire/BBCView cited source
  6. [6]

    In a February 26th statement, Amodei wrote that the company "cannot in good conscience accede" to the Pentagon's request for an "any lawful use" contract term.

    Reported Supported Source: Dario Amodei statementView cited source
  7. [7]

    Maven is operated by Palantir and organizes intelligence and other data for the Pentagon.

    Reported SupportedView cited source
  8. [8]

    Multiple people familiar with the department's use of AI told the BBC that Claude had become an important tool within Maven for processing large volumes of information; removing it was more complicated than terminating access to a standalone chatbot.

    Reported Supported Source: People familiar with the department's AI use, via BBCView cited source
  9. [9]

    Lauren Kahn, a senior research analyst at Georgetown University's Center for Security and Emerging Technology and a former defense official, told the BBC that Claude's integration made it difficult to remove quickly.

    Reported Supported Source: Lauren Kahn, via BBCView cited source
  10. [10]

    "Once they become integrated it can be painful to remove them."

    Reported Supported Source: Lauren Kahn, told the BBCView cited source
  11. [11]

    The BBC reports that OpenAI tools have gained wider adoption in some military departments.

    Reported Supported Source: BBC NewsView cited source
  12. [12]

    A DefenseScoop report said that, as of October 5th, the Pentagon had removed Anthropic models from its systems and was using or transitioning to models from Google, xAI and OpenAI.

    Reported Supported Source: DefenseScoopView cited source
  13. [13]

    Hegseth announced the supply-chain-risk designation against Anthropic on February 27th; the formal determination followed on March 3rd.

    Reported SupportedView cited source
  14. [14]

    On September 25th, the D.C. Circuit rejected Anthropic's challenge to the Pentagon's designation; Judge Karen LeCraft Henderson dissented.

    Reported SupportedView cited source
  15. [15]

    The D.C. Circuit majority concluded that the department could treat Anthropic's ability and willingness to restrict Claude's use as a supply-chain risk, and said the department acted over a contract term it considered essential to national security.

    Reported SupportedView cited source
  16. [16]

    The ruling did not settle the practical question of how quickly a deployed model can be removed once it is woven into operational software.

    Reported Supported Source: runtimewire analysis of BBC reportingView cited source
  17. [17]

    The Pentagon had awarded contracts to Anthropic, Google, OpenAI and xAI in July 2025, so replacing Claude did not inherently require a new contract.

    Reported SupportedView cited source
  18. [18]

    Some workflows did require migration and modification, according to the Defense Department's Chief Digital and Artificial Intelligence Office.

    Reported Supported Source: DoD Chief Digital and Artificial Intelligence OfficeView cited source
  19. [19]

    The available reporting does not establish how many users, deployments or contractor systems were affected by the phaseout.

    Reported SupportedView cited source
  20. [20]

    The department's statement does not specify when final access was removed or whether every contractor environment has completed the same process.

    Reported SupportedView cited source
  21. [21]

    Google, OpenAI and xAI, the providers DefenseScoop says the Pentagon is using or moving to, are the three July 2025 awardees other than Anthropic.

    Derived Supported

Sources

2 independent publishers whose own reporting we read for this story.

  1. runtimewire.com

    1 article · October 5, 2026

    Pentagon says it has stopped using Anthropic's Claude after a six-month phaseout
  2. techstrong.ai

    1 article · October 5, 2026

    Pentagon Cuts Ties with Anthropic Over AI Safety Standoff: Report - Techstrong.ai

Topics and entities

Topics

Entities