On Thursday, February 26 2026, OpenAI claimed to have the same limits as Anthropicon defense AI; no autonomous weapons, no domestic surveillance, no exceptions. But by Friday evening, OpenAI had signed a classified Pentagon contract. Meanwhile, Anthropic walked away from the same deal.
Public backlash hit fast. Employees learned about the deal from external outlets, not their CEO. Many users flooded the app store with one-star reviews and also appeared outside the OpenAI headquarters in San Francisco: “Where are your red lines?”.
Within days, OpenAI CEO Sam Altman openly admitted that the deal had been rushed. “We were genuinely trying to de-escalate things, but I think it just looked opportunistic and sloppy”, he said. But despite all of this, OpenAI kept the deal.
What the Pentagon Deal Says
The agreement sits on a stack of deals; a $200 million prototype agreement from June 2025, a deployment to the Department of Defense’s GenAI military classified networks and now, a classified deal.
OpenAI published an official announcement and claimed the company still has three major red lines; no autonomous weapons, no domestic surveillance and no social credit scoring.
Still, critics noticed a major loophole. OpenAI stated that AI will not control weapon systems unless the law requires a human to be in charge but nobody knows what that means in practice.
On top of that, the Pentagon never released the full contract. OpenAI later updated its terms to ban domestic surveillance, but Senator Ron Wyden said even that was not enough. Intelligence agencies already build detailed profiles using data bought from commercial sources. The update closes one door but leaves another wide open.
The Companies Winning the Defense AI Gold Rush
Right now, OpenAI just stepped into a market Palantir, Anduril and Google already dominate. Palantir brought in $4.48 billion in 2025, a 56% jump year over year. Anduril leads the counter-drone space. Google quietly returned to defense work after walking away in 2018.
Meanwhile, the global defense AI market stands at $28 billion today and tracks toward $65 billion by 2034. For all three companies, defense is the core of what they do and OpenAI simply decided to join.
Beyond the money, however, the Pentagon also made the stakes clear early on. When Anthropic walked away from talks, the Defense Department labeled it a supply-chain risk within hours, sending an unmistakable message to every AI company in America. Join or become an enemy.
Accountability at the Edge of Defense AI
Currently, there’s no public contract, no clear explanation of how the safety system works and a surveillance loophole that was caught within 48hrs. Furthermore, at the all-hands meeting, Altman confirmed that Defense Secretary Hegseth, not OpenAI engineers, holds final say over how the Pentagon uses the system.
Ultimately, OpenAI asks the public to trust a set of limits that nobody outside the company can check, enforced by an institution that just punished a competitor for asking for those same protections.
As OpenAI’s defense AI footprint grows from the Pentagon toward NATO, the tools to catch mistakes and change course remain almost entirely absent and the global stakes keep rising.
