Anthropic Draws a Line With the Pentagon

The AI-First Military Agenda with Anthropic

The Pentagon is moving aggressively to bring cutting-edge commercial AI into defence work.

A January 2026 military memorandum outlining the US military’s artificial intelligence strategy called for the country to become an AI-first fighting force and to accelerate the integration of leading AI models across warfighting, intelligence, and enterprise operations.

What stands out is that many of the most advanced AI capabilities are now concentrated in commercial firms rather than government labs.

The innovation cycle in venture-backed companies moves in months, while traditional defence procurement cycles often move in years.

Without access to commercial AI providers, the government would likely be slower, less adaptive, and far more expensive in building out these capabilities.

That said, there are still important ethical concerns, and Anthropic appears to be pushing back on some of them.

In particular, Anthropic wants assurances that its models will not be used to spy on Americans.

It also wants to ensure its models are not used for autonomous weapons.

What are the Best ASX Stocks to invest in right now?

Check our buy/sell tips

National Security vs Model Guardrails, The New AI Flashpoint

In July 2025, the Department of Defense awarded contracts worth up to US$200 million each to four companies, Anthropic, OpenAI, Google DeepMind, and xAI, to prototype frontier AI capabilities aligned with US national security priorities.

But the relationship with Anthropic appears to have broken down.

According to this version of events, Defense Secretary Pete Hegseth gave Anthropic an ultimatum to accept the government’s demands. When Anthropic refused, Hegseth reportedly said the company’s stance was fundamentally incompatible with American principles, and the Pentagon designated Anthropic a supply-chain risk to national security.

That response highlights the growing tension between national security priorities and the guardrails some AI companies want to maintain around how their models are used.

The Pentagon’s AI Push Runs Into a Civil Liberties Wall

We have also seen similar internal tension at OpenAI.

Following OpenAI’s controversial agreement with the Department of Defense, Caitlin Kalinowski resigned, saying the announcement was rushed and made without the guardrails being clearly defined.

OpenAI confirmed her departure and defended the Pentagon agreement, saying it includes safeguards and clearly defined red lines that prohibit domestic surveillance and autonomous weapons.

However, critics are still concerned that these safeguards are too vague.

The main issue is that unclear language can leave room for loopholes, especially around surveillance.

OpenAI’s critics argue that if the terms are not tightly defined, the government could still engage in forms of mass surveillance while claiming it remains within the rules.
For example, if the government buys bulk location data from a private company, does that count as unconstrained monitoring?

Most civil liberties groups would likely say yes.

The government, however, could argue no.

And that grey area is exactly why critics remain uneasy.