OpenAI Signed the Military Contract. Was There Ever Really a Choice?

2.5 million people quit ChatGPT. OpenAI employees publicly called it a betrayal. Sam Altman admitted the optics were bad. But here is the question the outrage mostly skipped: if the most powerful AI systems in the world are going to be used by militaries regardless, is refusing to be in the room actually the ethical position?

In late February 2026, the Trump administration demanded that Anthropic remove ethical restrictions prohibiting its Claude AI from being used for domestic mass surveillance and autonomous weapons. Anthropic's CEO Dario Amodei refused. The Pentagon designated Anthropic a national security supply chain risk and blacklisted it from all government systems.

Within hours, OpenAI announced a deal with the Department of Defense.

The backlash was fast, organized, and large. The #QuitGPT movement spread across social media. 2.5 million people canceled subscriptions, pledged to stop using ChatGPT, or shared the boycott publicly. ChatGPT uninstall rates jumped roughly 200%. Claude briefly became the number one free app on the Apple App Store. OpenAI employees published an open letter supporting Anthropic's position. One researcher, Leo Gao, publicly accused his employer of engaging in "window dressing" to make the deal appear more restricted than it was. Sam Altman held an X AMA, admitted "the optics don't look good," and called the execution sloppy.

All of this is accurate. What is also worth sitting with is the harder version of the question.

The Question the Outrage Mostly Skipped

Elon Musk was once asked in an interview whether he would board a plane if he knew it had some non-trivial probability of crashing. His answer, roughly paraphrased: it's not really a choice — if you don't get on, someone else does, and the plane flies anyway. The question isn't whether the flight happens. The question is who's in the cockpit.

Applied here: AI systems are going to be used by the US military. This is not a prediction. It is a description of what is already happening and will continue happening regardless of any individual company's decision. The Defense Department has been integrating AI tools across logistics, intelligence analysis, communications, and targeting support for years. The question is not whether military AI exists. The question is which AI, built by whom, with what constraints embedded in the architecture.

Anthropic drew a line and held it. They were blacklisted. OpenAI moved in and signed without those constraints. So the outcome of Anthropic's principled refusal was: Anthropic is out, OpenAI is in, and the Pentagon is using AI with fewer safeguards than Anthropic would have required. The ethical statement produced the least ethical outcome.

That is not an argument that Anthropic was wrong to refuse. It is an argument that the situation doesn't resolve cleanly into a position where anyone gets to feel good about it.

How the Industry Got Here

Google faced almost this exact situation in 2018 with Project Maven, an AI contract to analyze drone footage for the Department of Defense. Employees protested. Google ultimately did not renew the contract. The drone program continued — with different vendors. IBM has sold surveillance systems to governments. Amazon sold facial recognition to law enforcement. Microsoft has contracts with ICE. Every major technology company has at some point drawn a line, defended the line under pressure, and then quietly moved the line when the pressure changed or the revenue was large enough.

OpenAI was founded as a nonprofit in 2015 with an explicit mission: ensure artificial general intelligence benefits all of humanity. It converted to a capped-profit model in 2019, took $10 billion from Microsoft, and has spent the years since optimizing for commercial and government contracts. The Pentagon deal was not a betrayal of a principle that was being honored. It was the latest step in a trajectory that started when the nonprofit became a company.

Altman's admission that the deal was rushed and the execution sloppy is honest in one sense and evasive in another. The problem is not that it looked opportunistic. It is that the company that signed a classified defense contract hours after its competitor was punished for demanding safeguards had, in that moment, demonstrated exactly where it stood when values and contracts came into conflict.

The Real Problem Underneath All of This

The deeper issue isn't OpenAI specifically. It's that the AI industry has no coherent framework for what these systems should and shouldn't be used for — and neither does the government asking for them, nor the public responding to them.

"No autonomous weapons and no domestic surveillance" was Anthropic's stated line. OpenAI said it achieved equivalent restrictions through its deal, just differently. An OpenAI employee said publicly that the "all lawful purposes" language in the contract was window dressing. Altman didn't dispute that characterization with specifics. The actual restrictions in a classified defense contract are, by definition, not publicly verifiable. Both sides of this argument are making claims about a document you cannot read.

What is publicly verifiable is the pattern. AI labs repeatedly state principles about what their technology will not be used for. Government and commercial pressure then creates situations where those principles become inconvenient. The principles get revised, the contract gets signed, and the PR team issues a statement about responsible use and ongoing commitment to safety. This has happened enough times, across enough companies, that it is now a genre rather than an exception.

The people most alarmed about unsafe AI are building it. The companies most committed to safety are competing for defense contracts. The government demanding access is the same government with no regulatory framework for what that access enables. Everyone in this situation has a coherent individual explanation for their position. The system-level picture is considerably less coherent.

Where It Stands

OpenAI kept the Pentagon contract. Anthropic remains blacklisted from federal contracts. Claude became the number one app on the App Store for a week. ChatGPT retained over 60% market share. Altman apologized for the optics. No one changed the underlying situation.

The next version of this conversation is already being scheduled somewhere in Washington and San Francisco simultaneously. The specific companies and the specific contract will be different. The structural dynamic will be identical.

The outrage at OpenAI was justified. The assumption that a different choice would have produced a better outcome is harder to defend. These are not the same argument, and conflating them is how the real problem keeps not getting addressed.

Sources & verification

• NPR — Trump orders government to stop using Anthropic; Pentagon designates supply-chain risk; Anthropic vows legal challenge (Feb. 27, 2026)
• Fortune — OpenAI secures Pentagon deal hours after Anthropic blacklisting; deal guardrails detail (Feb. 28, 2026)
• Fortune — Altman all-hands; OpenAI red lines in contract; "edge systems" exclusion (Feb. 27, 2026)
• Fortune — Altman AMA; "optics don't look good"; employee open letter; Leo Gao criticism (March 2, 2026)
• Axios — Pentagon moved to OpenAI despite similar red lines; deal terms detail; Defense Production Act threat (March 4, 2026)
• Fortune — 30+ OpenAI and Google DeepMind employees file amicus brief supporting Anthropic; Jeff Dean involvement (March 10, 2026)

Article evidence and links checked Aug. 24, 2026. Sources updated to primary and institutional references. Editorial process →