Anthropic Refused Military Weapons and Got Banned. OpenAI Said Yes and Got the Contract. Who's Right?

In February 2026, Anthropic told the US Department of Defense that Claude could not be used for autonomous weapons systems or mass surveillance of American citizens. The Pentagon's response was swift and unprecedented: Defense Secretary Pete Hegseth designated Anthropic a "supply chain risk" — the first time that classification was applied to an American technology company. The designation required all defense contractors to verify they weren't using Claude for Pentagon-related projects, with a six-month window for complete removal.

Within weeks, OpenAI signed a contract with the Defense Department to fill the gap. The company that started as a nonprofit dedicated to ensuring AI "benefits all of humanity" chose to serve the military when its primary competitor chose not to. The contrast couldn't be starker: Anthropic refused military applications at enormous financial cost (hundreds of millions to potentially billions in lost revenue). OpenAI accepted them. The Pope invited Anthropic's co-founder to present an encyclical on AI ethics. The Pentagon invited OpenAI to replace Anthropic.

This isn't a minor policy difference between two technology companies. It's the defining ethical split in the AI industry, and the outcome will determine what AI is used for — and what it isn't — for decades to come.

What Each Company Actually Did

Anthropic's position, rooted in its founding charter, is that certain AI applications are too dangerous to enable regardless of the customer. The company's Constitutional AI framework — the document that shapes Claude's behavior — includes principles that conflict with unrestricted military use. When the Pentagon demanded unrestricted access, Anthropic offered restricted access with safeguards. The Pentagon considered the safeguards unacceptable and issued the supply chain risk designation.

Anthropic responded aggressively: the company sued the Trump administration in federal court, calling the designation "unprecedented and unlawful." A federal judge issued a preliminary injunction blocking enforcement, but the case remains active. The company estimated the ban put hundreds of millions to multiple billions of dollars of 2026 revenue at risk. Meanwhile, Anthropic's IPO preparations continue with a potential $900 billion valuation, suggesting investors don't view the military refusal as a business mistake — they view it as a brand asset.

OpenAI's position is more pragmatic. The company, now structured as a for-profit entity valued at $852 billion and preparing its own IPO, has consistently moved toward commercial relationships that maximize revenue and market position. Accepting the Pentagon contract filled a vacuum that Anthropic created — strategically logical from a business perspective. OpenAI has not publicly articulated an ethical framework for what military applications it would refuse, if any.

The irony deepens when you consider the origin stories. Anthropic was founded in 2021 by researchers who left OpenAI because they were concerned about the company's direction — specifically, what they perceived as insufficient attention to AI safety. Now the company they left is accepting the military contract they refused. The split that began over safety philosophy has materialized into concrete, consequential decisions about whether AI should be used to kill people.

The Arguments for Both Sides

The case for Anthropic's refusal: AI systems capable of autonomous lethal decision-making represent an unprecedented category of weapon — one that removes human judgment from the decision to end a human life. The potential for error (misidentification of targets), the potential for abuse (authoritarian governments using AI for suppression), and the irreversibility of lethal action all argue for categorical refusal. If AI companies normalize military AI applications, the technology will inevitably proliferate to regimes and contexts where oversight is minimal. The only way to prevent this is for frontier AI companies to establish norms of refusal before the technology is fully deployed. Anthropic is paying a short-term financial price for a long-term norm that, if adopted industry-wide, could prevent catastrophic applications.

The case for OpenAI's acceptance: AI is already being developed and deployed by authoritarian nations including China and Russia. If American AI companies refuse to work with the American military, the US loses its technological edge to adversaries who face no such restrictions. The responsible approach isn't refusal but engagement — working with the military to develop AI applications with appropriate safeguards rather than ceding the field to competitors with fewer ethical constraints. Additionally, defensive AI applications (cybersecurity, threat detection, logistics) are clearly beneficial and don't involve autonomous lethal decision-making. A blanket refusal fails to distinguish between offensive and defensive applications that have very different ethical profiles.

Both arguments have merit. The tension between them is genuine and unlikely to be resolved by either company's individual decision. What we're witnessing is the market determining whether ethical refusal or pragmatic engagement produces better long-term outcomes — financially, strategically, and morally.

What This Means for the People Who Use These AI Tools

If you use ChatGPT, you're using a product from a company that accepted a military contract for AI that another company refused. If you use Claude, you're using a product from a company that was banned by your government for refusing. Neither fact changes how the chatbot responds to your question about meal planning. But it shapes the ecosystem you're participating in, the values your subscription dollars support, and the future direction of AI development.

This is why understanding the companies behind AI tools matters beyond product features. Our ChatGPT vs Claude comparison covers product differences, and the AI privacy comparison examines data practices. But the military AI question adds a dimension that pure product comparisons miss: the ethical identity of the company building the tool you interact with daily. For anyone who wants to use AI tools effectively regardless of which company they choose, the free Prompt Optimizer works across all platforms, and TresPrompt brings optimization directly to your AI sidebar.

Frequently Asked Questions

Is Anthropic actually banned by the US government?

The Pentagon designated Anthropic a "supply chain risk," requiring defense agencies and contractors to phase out Claude usage within six months. However, a federal judge issued a preliminary injunction partially blocking enforcement. Simultaneously, the Pentagon is using Anthropic's Mythos model for cybersecurity through Project Glasswing. The situation is contradictory and evolving — Anthropic is technically banned for general defense use while being deployed for specific cybersecurity operations.

Did OpenAI always allow military use?

No — OpenAI updated its usage policies in January 2024 to remove a previous ban on "military and warfare" applications. The policy change occurred approximately one year before the company signed its Pentagon contract. The timing suggests a deliberate strategic shift to enable military partnerships that were previously prohibited under the company's own rules.

Does it matter which AI tool I use, ethically?

That depends on your values. If you believe AI companies should refuse military applications, choosing Claude supports that position financially. If you believe AI companies should work with democratic governments for defense purposes, choosing ChatGPT aligns with that view. If you're purely evaluating products on features and capability, the ethical differences may not affect your decision. There is no objectively "correct" answer — only choices that reflect individual values.

Will other AI companies take a position on military AI?

Google has the most ambiguous stance — the company faced internal protests over Project Maven (military AI contract) in 2018 and subsequently established AI principles that restrict certain applications. Meta, xAI, and smaller AI companies have not articulated clear positions on military use. As AI capabilities increase and government demand grows, every frontier AI company will face this question explicitly. The Anthropic-OpenAI divergence established the two poles; other companies will position themselves along the spectrum.

Has the Vatican's support helped Anthropic?

The Pope's encyclical presentation alongside Anthropic's co-founder provides powerful moral legitimacy for the company's stance. Whether this translates to commercial advantage is debatable — enterprise customers and investors may value the Vatican's implicit endorsement, while government customers clearly do not. The symbolic value is significant regardless: Anthropic's refusal is now associated with the moral authority of the Catholic Church, making political punishment more difficult to sustain without appearing to oppose human dignity itself.

Disclosure: Some links in this article are affiliate links. We only recommend tools we've personally tested and use regularly. See our full disclosure policy.