Pentagon Calls Anthropic an "Unacceptable Risk" as Tech Giants

Aerial view of the Pentagon building — Photo via Unsplash

TL;DR:

March 18: Pentagon filed a 40-page rebuttal arguing Anthropic's access to military systems "would introduce unacceptable risk into DoW supply chains"
The Pentagon's argument: Anthropic might disable or alter Claude mid-mission if the company believes its policies are being violated
Tech industry pushback: Microsoft, Google employees, OpenAI employees, and trade groups representing the entire tech sector filed amicus briefs backing Anthropic
150 retired federal and state judges also filed a brief warning the Pentagon's action misuses supply chain risk powers
Pentagon's backup plan: Building government-owned AI models. OpenAI and xAI approved for classified work while alternatives come online
Section 702 connection: The FISA surveillance authority expires in 32 days. The Pentagon wants AI that won't question surveillance orders

The Pentagon's March 18 Rebuttal

On March 18, the Department of Defense filed its first formal response to Anthropic's lawsuits: a 40-page document that escalated the rhetoric.

The filing warned that Anthropic's access to "technical and operational warfighting infrastructure would introduce unacceptable risk into DoW supply chains."

That's Pentagon-speak for: we don't trust this company near our systems.

The government's argument centers on a specific fear: what happens if Anthropic decides the military is violating its policies mid-operation?

From the filing: Anthropic could "attempt to disable or materially alter its foundation models" if the company believes its internal policies are being breached during wartime or other high-stakes missions. That possibility, the Pentagon argued, "undermines the reliability, availability, and command authority required for defense systems."

Translation: the Pentagon thinks Claude might refuse orders during combat. And they're not willing to risk it.

Tech Industry Lines Up Behind Anthropic

While the Pentagon escalated, the tech industry mobilized.

Microsoft filed an amicus brief in federal court in San Francisco backing Anthropic's request for a temporary restraining order. The brief was also signed by 22 retired military officials.

More than 30 Google and OpenAI employees filed their own brief, including Google DeepMind chief scientist Jeff Dean. These are workers at Anthropic's competitors (companies that stand to benefit if Anthropic loses military contracts) arguing the Pentagon's action threatens the entire AI industry.

The major tech trade groups joined in. The Computer & Communications Industry Association (CCIA), Information Technology Industry Council (ITI), Software & Information Industry Association (SIIA), and TechNet filed a joint brief representing essentially the entire technology sector.

And then there were the judges. Nearly 150 retired federal and state judges filed their own amicus brief warning that the Pentagon is misusing a tool designed for foreign adversaries.

The supply chain risk designation has previously been applied to Huawei and ZTE: Chinese companies with alleged backdoors for Beijing. Using it against an American company that simply wants ethics guardrails in writing? The judges called it a "category error with consequences that extend far beyond this dispute."

What the Pentagon Actually Wants

Strip away the legal language and the dispute comes down to two words: "unrestricted access."

Defense officials demanded that Anthropic grant the military access to Claude for "all lawful purposes." No contractual clauses limiting use cases. No company veto power over how AI gets deployed.

Anthropic wanted two things in writing:

No mass surveillance of Americans
No fully autonomous weapons (AI systems that select and engage targets without human approval)

The Pentagon's position: trust us, those uses are already illegal. We don't need it in a contract.

Anthropic's position: if it's already illegal, why not put it in writing?

The Pentagon didn't budge. Now we're in federal court.

The Pentagon's Plan B: Government-Owned AI

While fighting Anthropic in court, the Pentagon isn't waiting around.

On March 17, officials confirmed the military is actively building tools to replace Claude. Engineering work has begun on "multiple LLMs into government-owned environments." The Pentagon expects operational use "very soon."

In the meantime:

OpenAI and xAI have been approved for classified military work
Google's Gemini is being introduced across the Pentagon workforce

The strategy shift is significant. Rather than negotiate with AI companies that have ethics guidelines, the Pentagon wants to build and control its own AI. No safety teams. No usage policies. No company that might say no.

A defense official told Bloomberg: "Sovereign control over models, data, and guardrails will deliver speed with assurance."

Sovereign control. That's the tell. The Pentagon doesn't want partners with principles. It wants tools that follow orders.

The Surveillance Connection

This fight is playing out while Section 702 of FISA, the law authorizing warrantless surveillance of foreigners (which sweeps up millions of Americans' communications), has 32 days until it expires on April 20.

Anthropic's "red lines" include refusing AI use for mass surveillance of Americans. The Pentagon calls that an unacceptable risk.

Connect the dots.

Section 702 already allows backdoor searches of Americans' communications without a warrant. The Pentagon wants AI that won't blink when ordered to analyze that data. Anthropic built guardrails specifically to prevent this.

The timing isn't coincidence. This is a preview of the surveillance fights to come.

What Happens Now

Anthropic's lawsuits are in U.S. District Court for the Northern District of California. The company is seeking a temporary restraining order to pause the supply chain risk designation while the case proceeds.

If Anthropic wins, it sets precedent: AI companies can maintain ethics guardrails even for military contracts. If the Pentagon wins, it sends a message to every tech company: safety guidelines make you a liability.

The court's ruling could reshape who controls AI ethics: the companies that build these systems, or the government agencies that want to use them without restrictions.

With Section 702 about to sunset and the Pentagon actively building uncontrolled AI alternatives, the stakes go well beyond one company's contract.

Sources

Published: March 19, 2026