The Pentagon vs. Anthropic: A Standoff Over Military AI

A dispute over ethical guardrails for artificial intelligence has escalated into a federal ban, legal action, and a landmark court ruling.

The Core Conflict

The U.S. Department of Defense and artificial intelligence company Anthropic have been engaged in a public dispute over the terms governing the military's use of the company's Claude AI model. The conflict, which began in February 2025, centers on Anthropic's request for contractual assurances that its technology would not be used for mass surveillance of U.S. citizens or fully autonomous weapons.

The dispute has resulted in the Pentagon designating Anthropic as a "supply chain risk," a federal directive for agencies to cease using the company's products, and subsequent legal action from Anthropic. Several other major AI companies, including OpenAI, Google, and xAI, have reached agreements with the Pentagon for the use of their AI tools in classified military networks.

Background of the Dispute

Negotiations and Breakdown

Anthropic had previously deployed its Claude AI model across the DoD's classified networks under a contract valued at up to $200 million. The company was the first major AI lab approved for use on classified military systems.

"Domestic mass surveillance and fully autonomous weapons are uses simply outside the bounds of what today's technology can safely and reliably do." — Dario Amodei, Anthropic CEO

In early 2025, negotiations began between Anthropic and the Pentagon regarding the terms of a new or renewed contract. Anthropic CEO Dario Amodei stated that the company sought guarantees that its AI tools would not be used for two specific purposes: mass surveillance of American citizens and fully autonomous weapons systems that could make lethal decisions without human input. Amodei characterized these uses as "bright red lines" and stated they were beyond the safe and reliable capabilities of current technology.

The Pentagon, through Defense Secretary Pete Hegseth and other officials, maintained that private companies should not dictate how the military uses technology and demanded that Anthropic agree to "all lawful uses" of its products. Pentagon officials stated that the military had no intention of using AI for illegal mass surveillance or fully autonomous weapons, asserting that existing federal law and Pentagon policies already prohibit such uses.

On February 27, 2025, a deadline set by the Pentagon passed without an agreement. Amodei announced that new contract language from the Defense Department showed "minimal progress" in preventing Claude's use for the restricted applications, describing it as containing "legalese that would allow those safeguards to be disregarded at will."

Pentagon Ultimatums and Threats

During the negotiation period, Pentagon officials issued several warnings to Anthropic:

• Defense Secretary Hegseth stated he would terminate the $200 million contract if Anthropic did not comply.
• Officials threatened to designate Anthropic as a "supply chain risk to national security."
• Officials threatened to invoke the Defense Production Act, a law typically used during national emergencies, to compel Anthropic to provide its technology without restrictions.

Emil Michael, the Pentagon's Undersecretary for Research and Engineering, publicly criticized Amodei on social media, accusing him of having a "God-complex" and of lying.

Federal Actions Against Anthropic

Supply Chain Risk Designation

On March 6, 2025, the Pentagon formally designated Anthropic as a "supply chain risk." Pentagon Chief Information Officer Kirsten Davies issued a memorandum directing all senior military leadership to remove Anthropic's AI products from their systems within 180 days.

The memo stated that Anthropic's AI posed "an unacceptable supply chain risk for use in all Department of War systems and networks," including those for nuclear weapons, ballistic missile defense, and cyber warfare. The designation required any company contracting with the Pentagon to cease using Anthropic products on related work.

"The supply chain risk designation is likely both contrary to law and arbitrary and capricious." — Judge Rita F. Lin

Defense Secretary Hegseth stated that no entity doing business with the U.S. military could engage in commercial activity with Anthropic. This designation, typically reserved for foreign adversaries, marked the first known instance of the federal government applying it to a domestic American company.

Presidential Directive

President Donald Trump issued a statement on Truth Social directing all federal agencies to "IMMEDIATELY CEASE all use of Anthropic's technology." The directive included a six-month phase-out period for the Department of Defense and other agencies currently using the company's products. Trump stated that Anthropic made a "disastrous mistake" by attempting to dictate terms to the Department of War and affirmed that the government would not conduct future business with the company.

The Government Services Administration also terminated its contracts with Anthropic.

Other AI Companies' Agreements with the Pentagon

Following the breakdown of negotiations with Anthropic, several other major AI companies reached agreements with the Pentagon for the use of their AI tools in classified military networks.

OpenAI

OpenAI CEO Sam Altman announced an agreement with the Pentagon to deploy the company's AI models, including ChatGPT, within the DoD's classified network. Altman stated the agreement included:

• Prohibitions against domestic mass surveillance
• A requirement for human responsibility in the use of force, including for autonomous weapon systems
• Technical safeguards to ensure the models function as intended

Altman later acknowledged that the deal was announced hastily and that the company should not have rushed it. He announced revisions to the contract, including new language specifying that "the AI system shall not be intentionally used for domestic surveillance of U.S. persons and nationals." Altman also confirmed that the Defense Department agreed that OpenAI's tools would not be employed by intelligence agencies, including the NSA.

"People defending the United States should have the best tools in the world." — Sam Altman, OpenAI CEO

Altman publicly stated that OpenAI shares similar "red lines" with Anthropic and expressed hope that the Pentagon would offer Anthropic comparable terms. He criticized the Pentagon's threats against Anthropic and stated that Anthropic should not be designated as a supply chain risk.

Google, Microsoft, Amazon, Nvidia, and SpaceX

The Pentagon announced partnerships with Google, Microsoft, Amazon Web Services, Nvidia, and SpaceX to provide AI resources for classified computer networks. The initiative aims to augment warfighter decision-making and reduce the time required to identify and strike targets. Google's agreement includes non-binding language regarding domestic mass surveillance and autonomous weapons, but Google has no right to control lawful government operational decision-making. This contract represents a shift from Google's 2018 pledge to avoid lethal AI applications following employee protests over Project Maven.

xAI

Elon Musk's xAI signed a deal enabling its Grok model to operate within secure US military environments, aligning with the Defense Department's standard terms for lawful use. xAI received approval for use in classified settings.

Military Use of AI

Reported Deployment of Anthropic's Claude

Multiple reports indicated that the U.S. military utilized Anthropic's Claude AI model in active operations, including during a large-scale joint US-Israel military campaign against Iran. According to reports, military commands relied on Claude for intelligence analysis, support in target selection, and battlefield simulations linked to the strikes.

Claude was reportedly integrated into Palantir's Maven Smart System, which provided real-time targeting capabilities. Reports indicated that during the initial 24 hours of the escalating conflict, U.S. forces used AI targeting tools, incorporating Claude, to strike over 1,000 targets in Iran.

The use of Claude in military operations occurred despite Anthropic's ethics policy, which restricts using Claude for violent ends, weapons development, or surveillance without strict safeguards.

Pentagon Confirmation of AI Use

Admiral Brad Cooper, head of the US Central Command (CENTCOM), confirmed that the military uses a variety of AI tools in conflict situations. He stated that AI assists soldiers in processing large volumes of data and enables leaders to make informed decisions more rapidly, but humans retain the final authority for all targeting decisions.

Official Statements

Pentagon spokesperson Sean Parnell stated that the defense department has no interest in using AI for mass surveillance or developing autonomous weapons.

Anthropic's Policy Changes

Revised Safety Framework

Amid the dispute, Anthropic announced revisions to its core safety policy. The company adopted a nonbinding safety framework that it stated is subject to change. In a blog post, Anthropic indicated that its previous two-year-old Responsible Scaling Policy had shortcomings that could impede its ability to compete in the expanding AI market and noted it was inconsistent with the current anti-regulatory political climate in Washington.

The former policy mandated pausing the training of more powerful models if their capabilities exceeded the company's control and safety assurance. This specific measure was removed from the new policy. Anthropic argued that if responsible AI developers paused growth while other actors continued development, it could lead to a less safe global environment.

The new policy differentiates Anthropic's internal safety plans from its recommendations for the broader AI industry. The company acknowledged this framework is more flexible than its predecessor, describing them as "public goals" for which it will openly track progress, rather than "hard commitments."

Covert Performance Degradation Reversal

Anthropic released Claude Fable 5, a version of its AI model with additional safety guardrails. Initially, the company planned to covertly degrade the model's performance for users attempting to use Claude for frontier AI development without informing the user. Following backlash from the AI research community, Anthropic reversed this policy. The company now states that safeguards for AI development will be visible, alerting users if it suspects they are using Claude to build highly capable AI and either refusing the request or rerouting to a less capable model. Anthropic apologized for the initial approach.

Legal Actions

Anthropic's Lawsuits

Anthropic filed two federal lawsuits against the Trump administration:

1. Northern District of California: Filed against the Defense Department, Treasury, State, and Commerce Departments, along with their top officials. The lawsuit alleges that the government's actions violated Anthropic's First Amendment rights to free speech, exceeded the scope of supply chain risk law, and constituted illegal retaliation. Anthropic seeks to have the supply chain risk designation removed and related directives withdrawn. The company stated that the designation is jeopardizing "hundreds of millions of dollars in the near-term."

2. D.C. Circuit Court of Appeals: This suit challenges aspects of the government's authority, arguing that the Department of Defense exceeded its authority and that its actions were "arbitrary, capricious and an abuse of discretion."

"Nothing in the governing statute supports the Orwellian notion that an American company may be branded a potential adversary and saboteur of the U.S. for expressing disagreement with the government." — Judge Rita F. Lin

Anthropic CFO Krishna Rao stated in court filings that hundreds of millions of dollars in anticipated revenue from Pentagon-related work are at risk this year, with potential projected losses of up to $5 billion in sales if other companies are discouraged from partnering with Anthropic. Chief Commercial Officer Paul Smith reported that business partners are showing "deep distrust and a growing fear of associating with Anthropic," with some customers pausing negotiations, requesting escape clauses, or canceling meetings.

Preliminary Injunction Hearing

On March 18, 2025, U.S. District Judge Rita F. Lin heard arguments in San Francisco regarding Anthropic's request for a preliminary injunction.

Judge Lin's Preliminary Remarks: Before the hearing, Judge Lin delivered prepared remarks indicating her initial assessment. She stated that while the Department of War is free to cease using Claude and seek other vendors, the core issue was whether the government's subsequent actions violated the law. She noted that the government took three actions now subject to the lawsuit:

1. The president announced an immediate ban on all federal agencies from contracting with Anthropic.
2. Secretary Hegseth stated that companies doing business with the U.S. military must sever commercial relationships with Anthropic.
3. The Department of War labeled Anthropic a "supply chain risk."

Judge Lin expressed concern that these actions did not appear tailored to the stated national security concern and seemed punitive. She stated they appeared to be an attempt to "cripple Anthropic" and potentially punish the company for criticizing the government's contracting position publicly.

Government Arguments: Department of Justice attorney Eric Hamilton argued that the supply chain risk designation was justified due to concerns that Anthropic could "manipulate" its software or install a "kill switch" following tense negotiations. Hamilton argued that the actions were not retaliatory but were based on Anthropic's disagreement with the government over the AI model's potential uses.

Anthropic Arguments: Anthropic's lawyer, Michael Mongan, argued that a company posing such a risk would likely not engage in a public dispute but would instead accept terms to facilitate malicious intent.

Temporary Block

On March 21, 2025, Judge Lin issued a preliminary injunction temporarily blocking the Pentagon's classification of Anthropic as a supply chain risk and halting the presidential directive for federal agencies to cease using Anthropic's technology.

In a 43-page ruling, Judge Lin wrote:

• The supply chain risk designation is "likely both contrary to law and arbitrary and capricious."
• "Nothing in the governing statute supports the Orwellian notion that an American company may be branded a potential adversary and saboteur of the U.S. for expressing disagreement with the government."
• The measures did not appear to be directed at the government's stated national security interests and instead "appear designed to punish Anthropic."
• Department of War records indicated Anthropic's designation was partly due to its "hostile manner through the press," which Judge Lin described as "classic illegal First Amendment retaliation."
• Anthropic's due process rights were likely violated, as the company lacked an opportunity to respond.
• The government's actions were "arbitrary and capricious," noting cordial contract negotiation emails between Pentagon officials and Amodei even as the military depicted Anthropic as a threat.

She also rejected a social media post by Hegseth that seemed to require military contractors to halt all "commercial activity" with Anthropic, stating it illegally extended to non-military work.

Judge Lin clarified that her ruling does not prevent the Trump administration from taking "lawful actions" such as choosing an alternative AI provider. She stayed her order for seven days to allow the government an opportunity to appeal.

Industry Support for Anthropic

Over 30 researchers from OpenAI and Google, including Google DeepMind's chief scientist Jeff Dean, filed an amicus brief supporting Anthropic in its legal dispute. They argued that the Pentagon's designation could negatively affect the US AI industry's competitiveness.

OpenAI CEO Sam Altman stated on social media that enforcing the supply chain risk designation would be detrimental to the industry and the country.

Organizations including Microsoft, the ACLU, and retired military leaders also filed amicus briefs supporting Anthropic. Nearly 500 employees from OpenAI and Google signed an open letter in support of Anthropic.

The Information Technology Industry Council (ITI), whose members include Nvidia, Amazon, and Apple, sent a letter to Defense Secretary Hegseth expressing concerns that the designation could create uncertainty and threaten the US military's access to high-quality products and services.

Continued Negotiations

Despite the legal actions, reports indicated that Amodei and Emil Michael resumed discussions in an attempt to reach an agreement. In an internal memo, Amodei reportedly indicated that near the end of negotiations, the Defense Department had offered to accept Anthropic's terms if they removed a "specific phrase about 'analysis of bulk acquired data,'" which Amodei stated aligned with a scenario the company aimed to prevent.

Public and Industry Reactions

Employee Protests

More than 600 Google employees signed an open letter to CEO Sundar Pichai opposing the use of the company's AI for classified Pentagon operations. Despite the protest, Google proceeded with the contract. Workers at Google DeepMind voted to unionize in April 2025, citing concerns over military AI contracts as a motivating factor.

Demonstrators gathered at OpenAI's headquarters in San Francisco to protest the company's Pentagon deal.

Public Support for Anthropic

Following the dispute, the Claude application experienced increased downloads, reaching top positions on app stores ahead of ChatGPT. Online forums featured discussions and calls for users to switch from ChatGPT to Claude in response to the agreement.

Congressional Reactions

• Senator Mark Warner (D-VA): Expressed deep disturbance over reports of the Pentagon attempting to "bully a leading U.S. company," emphasizing the need for strong AI governance mechanisms in national security contexts. He also criticized Trump's directive, raising concerns about whether national security decisions are driven by careful analysis or political considerations.

• Senator Thom Tillis (R-NC): Criticized the Pentagon's public handling of the dispute, advising that discussions with strategic vendors should occur privately.

• Senator Kirsten Gillibrand (D-NY): Described the supply chain risk designation as a "dangerous misuse" of authority that could negatively impact the U.S. AI sector.

Expert Commentary

• Jerry McGinn, director of the Center for the Industrial Base at CSIS, described the standoff as unusual for Pentagon contracting, noting that contractors typically do not dictate how their products are used. He attributed the public nature of the dispute to the novel and largely untested nature of AI technology.

• Amos Toh, senior counsel at the Brennan Center, called for greater AI oversight or regulation by Congress, noting that the law is not keeping pace with the rapid evolution of AI technology.

• Retired Air Force Gen. Jack Shanahan expressed sympathy for Anthropic's position, calling their red lines "reasonable" and noting that AI large language models are not yet suitable for fully autonomous weapons in national security contexts.

• Geoffrey Gertz, senior fellow at the Center for a New American Security, noted that designating Anthropic as a "supply chain risk" would be an unusual step, traditionally reserved for foreign adversary technology.

• Neil Chilson of the Abundance Institute labeled the decision a "massive overreach."

Key Statements from Officials

• Defense Secretary Pete Hegseth: Stated that companies must allow for any uses the Pentagon deems lawful and that the military's "AI will not be woke." He referred to Anthropic's stance as "woke AI."

• Emil Michael, Pentagon CTO: Stated it would have been irresponsible to rely on only one company, prompting a multi-provider approach. He stated that a company with differing policy preferences embedded in its model could "pollute the supply chain."

• Pentagon spokesperson Sean Parnell: Stated the military has no intention of using AI for illegal mass surveillance or autonomous weapons without human involvement, but emphasized that the Pentagon would not allow any company to dictate its operational decisions.

• OpenAI CEO Sam Altman: Stated that "people defending the United States should have the best tools in the world" and that OpenAI shares similar "red lines" with Anthropic regarding military AI use.

• Anthropic CEO Dario Amodei: Stated that domestic mass surveillance and fully autonomous weapons are uses "simply outside the bounds of what today's technology can safely and reliably do." He affirmed the company's support for all lawful uses of AI for national security, with the two narrow exceptions.