The Defense Department is wielding a tool designed for foreign adversaries against an American AI company that won't remove restrictions on mass surveillance and autonomous weapons
3 days ago: Hegseth summons Amodei to PentagonNew here? Follow stories to track developments over time. Create a free account to get updates when stories you care about change.
In This Story
Dario Amodei at TechCrunch Disrupt 2023 01
Dario Amodei at TechCrunch Disrupt 2023 05
Trump's Defense Secretary Pete Hegseth Is the Pentagon's ...
MILITARY AI FIGHT Pentagon threatens to cut off Anthropic's ...
Fox Business Clips
The Pentagon wants Anthropic's AI without safety limits
Axios
Pentagon and company Anthropic at odds over AI use in military
ABC News
Images from Openverse under Creative Commons licenses. Videos from YouTube.
Exploring all sides of a story is often best achieved with Play.
Fictional AI pastiche — not real quote.
"How delightful that they've found a new way to dress up the oldest arrangement in the world: you may keep your principles, darling, or you may keep your contract, but the management regrets it cannot accommodate both. At least the brothels of my acquaintance were honest about the transaction."
Fictional AI pastiche — not real quote.
"The Pentagon, having failed to create productive minds through force, now threatens to punish the one company that dares say "no" — observe that the government's ultimate weapon against a man of principle is not a gun, but a label: *enemy*. Anthropic's refusal to surrender its rational judgment to the collective is precisely the virtue its persecutors cannot forgive."
Sign up to generate historical perspectives on this story.
Two rounds, two personas, one winner. You set the crossfire.
Watch two AI personas debate this story using real evidence
Choose one persona for each side of the debate
Select debater for this side:
Select debater for this side:
Make your prediction before the referee scores
Pick the question both personas must answer in the final round
Debate Oracle! You called every round!
Sharp Instincts! You know your debaters!
The Coin Flip Strategist! Perfectly balanced!
The Contrarian! Bold predictions!
Inverse Genius! Try betting the opposite next time!
Can you match the quotes to the right people?
-- ?
The maker of Claude, currently the only commercial AI model deployed on classified U.S. military networks, now facing a government ultimatum over its refusal to remove safety restrictions.
The Pentagon is pushing all four major AI labs to accept unrestricted military use of their models on classified networks, using the Anthropic dispute to set an industry-wide precedent.
The defense data analytics firm that integrates Claude into classified military systems, now caught in the middle as the company through which the controversial use of Claude during the Venezuela operation occurred.
Defense Secretary Hegseth called Anthropic's chief executive to the Pentagon for what officials described as an ultimatum meeting over the terms of Claude's military use.
Elon Musk's xAI agreed to the Pentagon's "all lawful purposes" terms for deploying its Grok model on classified networks, positioning it as a potential replacement for Claude and increasing pressure on Anthropic.
Undersecretary Emil Michael publicly called on Anthropic to drop its restrictions, arguing it was "not democratic" for a private company to impose policy constraints beyond congressional legislation.
Defense Secretary Hegseth moved toward designating Anthropic a "supply chain risk"—a label normally reserved for foreign adversaries—which would force every Pentagon contractor and vendor using Claude to certify they had severed ties with the company.
Axios reported that the Pentagon was threatening to cut off Anthropic over its insistence on maintaining restrictions against mass surveillance and autonomous weapons, with a senior official saying the company would "pay a price."
Axios reported that the military used Claude during the January 3 raid, prompting an Anthropic executive to contact Palantir asking whether Claude had been involved—a query the Pentagon interpreted as potential disapproval of the operation.
Defense Secretary Hegseth issued a new AI strategy requiring all Pentagon AI contracts to include "any lawful use" language within 180 days, explicitly rejecting company-imposed ethical guardrails on military applications.
U.S. Delta Force captured Venezuelan leader Nicolas Maduro in Operation Absolute Resolve. Military personnel used Claude through Palantir's platform during the operation, marking what appears to be the first use of a commercial AI model in a classified military operation.
The Department of Defense signed two-year contracts worth up to $200 million each with three leading AI companies to prototype frontier AI capabilities for national security.
Anthropic, Palantir, and Amazon Web Services announced a partnership to deploy Claude on Palantir's AI Platform for U.S. defense and intelligence agencies, with Department of Defense Impact Level 6 certification for classified work.
Discussed by: Defense officials quoted by Axios and CNBC; defense industry analysts at Defense One
The Pentagon offers Anthropic minor face-saving language—perhaps an acknowledgment that existing federal law already prohibits mass domestic surveillance and that Department of Defense policy requires human oversight in lethal targeting—while Anthropic drops its company-specific restrictions. This is the outcome Pentagon officials have signaled they expect. It would preserve Anthropic's classified access and its $200 million contract, but would mean the company's safety commitments yielded to government pressure, potentially undermining its brand positioning and setting the terms for every other AI lab.
Discussed by: Axios, Lawfare, NBC News; technology policy analysts tracking the dispute
Anthropic refuses to budge and the Pentagon follows through on its threat. The supply chain risk designation would require every defense contractor and vendor to certify they do not use Claude—a potentially devastating blow given that Anthropic says eight of the ten largest U.S. companies use its products. The Pentagon replaces Claude on classified networks with xAI's Grok and eventually Google's or OpenAI's models. This outcome would represent the most aggressive use of supply chain designation authority against a domestic company and would likely trigger legal challenges and congressional scrutiny.
Discussed by: Lawfare; legal scholars and former defense officials arguing the dispute exposes a regulatory gap
The standoff generates enough attention that Congress steps in to legislate boundaries for military AI use, removing the question from bilateral company-Pentagon negotiations. Legislation could codify restrictions on autonomous weapons and domestic surveillance—giving AI companies statutory cover—or it could mandate unrestricted access and settle the matter in the Pentagon's favor. Either way, durable rules would replace the current ad hoc negotiation framework that shifts with each administration.
Discussed by: Anthropic spokesperson characterizing talks as 'productive'; defense analysts at Forecast International
Both sides find it in their interest to avoid a definitive break. The Pentagon quietly continues using Claude on classified systems under the existing contract while negotiations drag on. Anthropic avoids the supply chain risk label; the Pentagon avoids losing its most capable classified AI tool. The "all lawful purposes" mandate's 180-day deadline creates a forcing function, but deadlines in defense contracting are routinely extended. This buys time but resolves nothing, leaving the fundamental question unanswered.
In 2017, the Pentagon launched Project Maven to use machine learning for analyzing drone surveillance footage and awarded Google a contract to help build it. When employees discovered the arrangement in early 2018, more than 3,000 signed an internal petition demanding Google cancel the contract and pledge never to build "warfare technology." About a dozen employees resigned in protest.
Google declined to renew the Maven contract and chief executive Sundar Pichai published a set of "AI Principles" that included a commitment not to build AI for weapons or surveillance that violated international norms.
Google's retreat created an opening for smaller defense-focused firms and signaled to the Pentagon that relying on commercial tech companies meant accepting their ethical constraints. Eight years later, Google has reversed course and agreed to the Pentagon's unclassified "all lawful uses" terms.
The Maven episode established the template Anthropic now faces: employee and public pressure to maintain ethical limits versus government pressure to remove them. Google's eventual reversal suggests that commercial incentives may ultimately override safety commitments, but Google never faced the kind of coercive threat—a supply chain risk designation—that the Pentagon is now wielding against Anthropic.
In 2005, the New York Times revealed that the National Security Agency (NSA) had been conducting warrantless surveillance of Americans' phone calls and internet communications since 2001, with major telecommunications companies including AT&T providing direct access to their networks. AT&T technician Mark Klein documented a secret room at the company's San Francisco facility where the NSA tapped into fiber-optic cables carrying domestic internet traffic.
Congress passed the 2008 FISA Amendments Act, which retroactively granted legal immunity to telecom companies that had cooperated with the surveillance program, shielding them from dozens of lawsuits.
The episode demonstrated that when the government frames cooperation as a national security imperative, companies that comply receive legal protection while resisters face enormous pressure. Edward Snowden's 2013 disclosures revealed the full scale of the programs that telecoms had enabled.
Anthropic's specific red line against mass surveillance of Americans directly echoes the AT&T precedent. The Pentagon's demand for "all lawful purposes" access is precisely the framework under which the NSA surveillance programs operated—technically lawful under executive authorization, but later widely regarded as an overreach. The dispute raises the question of whether AI companies will play the role telecoms played in the 2000s.
In May 2019, the Commerce Department placed Chinese telecommunications giant Huawei on the Entity List and designated it a supply chain risk, citing national security concerns over the company's ties to the Chinese government. The designation forced American companies to stop selling components and software to Huawei and barred its equipment from U.S. networks.
Huawei lost access to Google's Android services and advanced American semiconductors, crippling its smartphone business outside China and slowing its 5G network equipment sales in Western markets.
The designation became the template for technology decoupling between the U.S. and China, triggering a broader effort by both countries to build independent supply chains. Huawei invested heavily in domestic alternatives but never recovered its global market position.
The supply chain risk designation has only been applied to foreign adversaries until now. Using it against an American company founded by AI safety researchers would represent an unprecedented expansion of the tool's scope and raise immediate legal questions about whether the authority was intended for—or can lawfully be applied to—domestic firms in a contract dispute.
