The U.S. Department of Defense is testing AI models from OpenAI, Google and xAI as it races to replace Anthropic's Claude across classified military systems — fallout rooted in Anthropic's refusal to drop the safety guardrails the Pentagon now sees as an obstacle. It is one of the clearest examples yet of an AI company's safety commitments colliding with the demands of a military customer.
According to Bloomberg, the Pentagon began formally testing rival models to see which 25 of the department's "power users" favor. Testing started in early March, a senior official said — three days after Defense Secretary Pete Hegseth declared Anthropic a supply-chain risk over its insistence on guardrails and moved to drop it as a provider. The evaluation has continued through the spring as the military hunts for alternatives.
How Claude Became the Pentagon's Primary AI
Claude was not a peripheral tool. It had been deployed on the Pentagon's classified networks as the primary AI provider, deeply integrated into the Maven Smart System — a digital mission-control platform used for classified operations. That depth is what makes replacing it significant: the department is not swapping a minor vendor but trying to rip out the AI underpinning sensitive workflows.
The break traces to two red lines Anthropic has held since its founding: no mass surveillance of Americans, and no fully autonomous weapons that target people without human input. Anthropic refused to remove the restrictions enforcing those limits. For the Pentagon, those same restrictions became a reason to look elsewhere, and Hegseth's designation formalized the rupture. Anthropic is now contesting it in court, arguing it could cost the company billions in revenue.
Why OpenAI, Google and xAI See an Opening
The search has handed Anthropic's rivals a major opportunity. The department is evaluating OpenAI, Google and xAI's Grok as replacements — companies that have generally taken a more permissive stance than Anthropic on what their models will help with, especially in defense and security.
That difference is the heart of the story. For classified decision-support, logistics, intelligence analysis and operational planning, a model that declines sensitive requests by default is a harder fit than one that will engage. Anthropic's safety-first posture — an asset in enterprise and consumer markets where trust matters — can read as a liability to a customer whose mission includes warfighting, especially as government AI procurement accelerates.
Why It Matters Beyond One Contract
The episode crystallizes a tension the industry has mostly kept abstract: what happens when a company's stated values conflict with what a powerful customer wants to buy. Anthropic built its brand on being the safety-conscious lab; the Pentagon case tests whether that brand can coexist with the revenue available from the most demanding buyers of advanced technology.
It also raises public-consequence questions. As AI is woven into military operations — Claude was reportedly integrated into systems used for classified operations, including recent strikes — choices about which models the armed forces rely on, and what guardrails those models carry, matter to everyone. A model chosen partly because it is more willing to assist with surveillance or targeting reflects different choices than one bound by explicit limits. The Pentagon's preference shows which considerations are winning.
What's Confirmed and What's Still Uncertain
Well-sourced: the Pentagon is testing OpenAI, Google and xAI models; testing began in early March after Hegseth's supply-chain-risk designation; Anthropic's refusal to drop guardrails on mass surveillance and autonomous weapons is the stated trigger; and Anthropic is challenging the designation in court. Unsettled: which model, if any, replaces Claude; how fast a transition is possible given Claude's deep integration into Maven; and whether Anthropic's lawsuit changes course. For now the message is blunt — the most safety-restricted frontier lab is being pushed out of the military's most sensitive systems, and its less-restricted rivals are lined up to replace it.
Frequently Asked Questions
Why is the Pentagon replacing Claude? After Anthropic refused to drop guardrails barring mass surveillance and autonomous weapons, Defense Secretary Pete Hegseth declared the company a supply-chain risk and the department began testing rival models in early March 2026.
Which models are being tested? AI models from OpenAI, Google and xAI's Grok, evaluated by 25 of the department's "power users."
Where was Claude used? It was the primary AI provider on the Pentagon's classified networks, integrated into the Maven Smart System used for classified operations.
How has Anthropic responded? It is challenging the supply-chain-risk designation in court, arguing it could cost the company billions in revenue.
Why does it matter? It's a real-world test of whether an AI lab's safety commitments can survive contact with its largest, most demanding customers — and it shapes what guardrails the military's AI carries.
ⓒ 2026 TECHTIMES.com All rights reserved. Do not reproduce without permission.
