Warren: Pentagon Coercing AI Firms Into Autonomous Weapons Work

Senator Elizabeth Warren sent formal letters to Defense Secretary Pete Hegseth and OpenAI CEO Sam Altman on Monday, calling the Pentagon's decision to blacklist Anthropic an action that "appears to be retaliation" and demanding answers about the terms of both companies' government contracts.

As reported by CNBC, Warren told Hegseth directly that the Department of Defense "could have chosen to terminate its contract with Anthropic or continued using its technology in unclassified systems" — framing the supply chain risk designation as a disproportionate and politically motivated response rather than a genuine national security determination.

"I am particularly concerned that the DoD is trying to strong-arm American companies into providing the Department with the tools to spy on American citizens and deploy fully autonomous weapons without adequate safeguards," Warren wrote in the letter.

A preliminary hearing in Anthropic's lawsuit against the Trump administration is scheduled for Tuesday in the U.S. District Court for the Northern District of California.

How We Got Here

The dispute traces back to a straightforward but consequential disagreement over contract language. Before the confrontation became public, Anthropic wanted explicit contractual bans on two specific use cases for its Claude models: fully autonomous weapons — systems capable of lethal action without human oversight — and domestic mass surveillance. The DOD, according to a senior department official who spoke to CNBC, wanted "unfettered access to Claude across all lawful purposes" and would not accept a vendor inserting itself into the chain of command by restricting how the military uses a capability.

Negotiations collapsed. On February 27, Defense Secretary Pete Hegseth posted on social media that he was directing the DOD to label Anthropic a "Supply-Chain Risk to National Security." President Trump directed all federal agencies to "immediately cease" use of Anthropic's technology. Trump told Politico: "Anthropic is in trouble because I fired them like dogs."

The formal supply chain risk designation followed a week later — a label that has historically been reserved for foreign adversaries, not American companies. It requires defense vendors and contractors to certify they don't use Claude in any work with the Pentagon, effectively trying to excise Anthropic from the broader defense contractor ecosystem.

The extraordinary nature of the designation wasn't lost on the defense and policy community. A letter to Congress signed by former defense officials, reported by CNBC, called it a "dangerous precedent," arguing that "blacklisting one of America's leading AI companies — and requiring its thousands of contractors and partners to sever ties as well — does not strengthen our competitive position."

The OpenAI Variable

Hours after Anthropic was blacklisted, OpenAI announced a deal with the DOD. The timing was not subtle. Sam Altman wrote publicly that OpenAI's contract includes "prohibitions on domestic mass surveillance and human responsibility for the use of force, including for autonomous weapon systems" — the same two limits Anthropic had demanded. The difference, Altman argued, was that OpenAI was confident existing laws, its "safety stack," and contract language were sufficient. Anthropic wanted explicit contractual guarantees. The DOD accepted OpenAI's framing and rejected Anthropic's.

Warren's letter to Altman probes exactly that gap. "Ultimately, it is impossible to assess any safeguards and prohibitions that may exist in OpenAI's agreement with DoD without seeing the full contract, which neither DoD nor OpenAI have made available," she wrote. Business Insider later reported that OpenAI is amending its contract amid ongoing backlash, though terms of those amendments haven't been disclosed either.

Where this leaves the public: two AI companies, both claiming they have protections against autonomous weapons and mass surveillance in their government contracts, but neither willing to publish the contract language that would allow anyone to verify those claims.

The First Amendment Argument

Anthropic's lawsuit, filed March 9 in the Northern District of California, goes beyond contract disputes. "The Constitution does not allow the government to wield its enormous power to punish a company for its protected speech," the filing states, as reported by Wired. The company also separately petitioned the U.S. Court of Appeals in Washington D.C., arguing the designation was "a pretextual form of retaliation in violation of the First and Fifth Amendments."

Anthropic's CFO Krishna Rao stated in a court filing that the government's actions could reduce the company's 2026 revenue by "multiple billions of dollars" — a number that reflects not just lost government contracts but the downstream effect of requiring the entire defense contractor supply chain to certify non-use of Claude.

The Iran context adds another layer. Even as the DOD was formally designating Anthropic a supply chain risk, CNBC has previously reported that the Pentagon was simultaneously using Claude models to support U.S. military operations in the ongoing conflict in Iran. That the government was both blacklisting and actively using the company's technology at the same time is the factual core of Warren's retaliation argument.

What the Tuesday Hearing Means

The preliminary hearing is a procedural step — Anthropic is seeking a stay on the supply chain designation while the larger case unfolds. Courts have historically been reluctant to second-guess executive branch national security determinations, which makes Anthropic's First Amendment framing strategically significant: it shifts the legal question from "did the DOD make a reasonable security call" to "did the government punish a company for refusing to remove speech-adjacent safety limits."

Analysis: Warren's letters don't change the immediate legal or political calculus — Senate Democrats lack the votes to force action with Republicans controlling both chambers. But they add a formal congressional record to a confrontation that has so far played out largely through court filings and social media. If the Tuesday hearing produces a stay, or if the full contract terms ever become public, this story has more distance to run. The underlying question — whether an AI company can impose usage limits on a government customer, or whether accepting government contracts means ceding that control entirely — is not going to be resolved by this case alone.